Politika, Politológia | Konzervativizmus » Kee-Yi - Conservatism, A Measurement Maze

Source: http://www.doksinet Conservatism – A Measurement Maze Ho Yew Kee bizhoyk@nus.edusg Yuan Yi yuanyi.ws@gmailcom Department of Accounting NUS Business National University of Singapore 15 Kent Ridge Drive Singapore 119245 Source: http://www.doksinet Conservatism – A Measurement Maze Abstract Conservatism has been a bedrock of accounting as it governs the prudent preparation of financial statements to prevent over‐optimism in financial reporting by management. However conservatism is like sunlight where one can feel the effects and warmth of it but has difficulty quantifying or measuring it. The purpose of this study is to investigate the robustness of accounting conservatism measures in their applications to different industries and conservatism impact assessment studies. Current conservatism research often adopts different conservatism measurement models arbitrarily and applies them to cross‐industry samples without controlling for industry differences. There are

also significant disagreements among findings of conservatism impact assessment studies. This study, through empirical assessment of an extensive sample of 43,434 firm‐years over a 23‐year period spanning 1988‐2010, aims to provide evidence to show the significant inconsistencies among five popular conservatism measurement models. Using Penman and Zhang’s (2002) earnings persistence regression and different unconditional conservatism measurement models, the conclusion reached is that Penman and Zhang’s model does not hold for all industries. Therefore, this study raises concerns about the arbitrary application of conservatism measurement models in current research and the reliability of results produced by such studies. Keywords: Conservatism, accruals, measurement, earnings persistence 1 Source: http://www.doksinet Conservatism – A Measurement Maze 1 Introduction Conservatism is one of the fundamental principles in accounting. Scholars have suggested that its

influence on accounting convention is both historical and entrenched (Watts, 2003a, 2003b; Basu, 1997, Sterling, 1970). The wide application of conservative accounting practices before the 1900s shows that they were generally accepted as positive practices that improve accounting information quality.1 However, in recent years, scholars and practitioners have begun to doubt the unquestioned application of accounting conservatism, especially when there is no conclusive research evidence to substantiate the benefits of conservative accounting practices. The regulators’ stand on conservatism has also begun to waver, and in recent years has become less supportive. For instance, Financial Accounting Standards Board (FASB) has revised its position on the quality of financial reporting by stating: “Understating assets or overstating liabilities in one period frequently leads to overstating financial performance in later periodsa result that cannot be described as prudent or neutral.”

FASB 8, September 2010 Although FASB’s position has shifted to support neutral financial reporting instead of conservative financial reporting, nonetheless conservatism is still an This is consistent with Robert Sterling’s observation in 1970 that in the presence of “effervescent optimism of the entrepreneur universal tendency to overvalue the enterprise”, the accountant set out in “solidarity and stability to combat overstatement, he proposed understatement, perhaps with the hope of striking a balance”. (Sterling, 1970, p256) 1 2 Source: http://www.doksinet entrenched concept. Some scholars have argued that accounting conservatism is beneficial in reducing potential regulatory, litigation, contracting costs, and taxation expenses (Watts, 2003a). Others, however, claim that conservatism is not desirable as it decreases the quality of accounting information or increases the abilities of firms to manage their earnings (Jackson and Liu, 2010). The controversy

surrounding conservatism points to the need to study the impacts of conservatism both in theory and empirically. This would then facilitate evaluation of the costs and benefits of accounting conservatism particularly in its role in accounting information provision. In 1993, Ross Watts raised the awareness of research in conservatism and proposed various research agendas2. This led to a boom in research on conservatism Many studies on the measures and impact assessment of accounting conservatism have been published since. Despite extensive studies conducted in the past two decades, the findings regarding the pros and cons of accounting conservatism are more anecdotal than conclusive. There is also limited applicability of these research findings in accounting regulations. Therefore the justification for excluding conservatism from accounting standards (FASB, 2010) so far is solely qualitative without strong research backing. In addition, anecdotal evidence shows inconsistencies in

results produced by studies adopting different conservatism measures (Chandra, 2011)3. Thus, quantitative evidence may be the missing link in the formulation of accounting policies and regulations with respect to accounting conservatism. Before scholars can assess the impact of conservatism, there needs to be a commonly recognized definition of accounting conservatism and methodology to measure accounting conservatism. Currently, there is neither a single authoritative 2 A Proposal for Research on Conservatism at the American Accounting Association (AAA) Convention. fact, Wang et. Al (2009) document convergent validity issues amongst the five measures 3In 3 Source: http://www.doksinet definition of conservatism nor a unanimously accepted conservatism measure. Wang et al. (2009) report that 94% of published papers adopt one or more of the five conservatism measures or their adaptations. The choice and application of these conservatism measurement models, however, is often

arbitrary. It is possible that the different conservatism measurements are highly correlated and proxies for the underlying accounting conservatism of a firm. However, this is only a conjecture Hence, the research results produced using different conservatism measures may not be comparable or reconcilable and may often lead to inconsistent results. This study raises, and seeks to provide an answer to the question “are conservatism measures too varied to be consistent?”. Our study serves to raise awareness of the severity of such inconsistencies, and to spur research interest in this area so that future conservatism research will consider the importance of the conservatism measures in terms of reliability, consistency and conclusiveness. This study extends the body of literature of conservatism research by systematically assessing the robustness of different conservatism measures in their application to different industries and on studies of accounting conservatism. Through

empirical analysis of extensive data comprising 43,434 firm‐years over an extended period of 1988‐2010, tests conducted in this study provide evidence to reveal the presence, extent, and impact of inconsistencies among conservatism measures. This study is organized as follows: Section 2 reviews major papers and prior research conducted in the area of accounting conservatism. Section 3 articulates the various hypotheses developed in this study while Section 4 provides the details of research design, methodology, sample data and results. Finally Section 5 presents the implications of this study and areas for future research. 4 Source: http://www.doksinet 2. Literature Review One of the primary reasons for the interest in accounting conservatism is its impact on quality of earnings which is also referred to as the informativeness of earnings.4 The informativeness of earnings is often measured by the ability of current earnings to provide indications of future earnings (Penman

and Zhang, 2002) 5. A general conclusion of the literature is that there are disagreements as to whether conservatism has negative effects on the persistence and predictability of earnings (Ruch and Taylor, 2011). For example, Penman and Zhang (2002) show that unconditional conservatism creates ‘hidden reserves’ which increase biases and error in current reported earnings, and hence reduces the ability of investors to predict future earnings. Adopting the negative accruals (NA) model by Givoly and Hayn (2000), Kim and Kross (2005), however, obtain results which suggest that increasing the level of conservatism increases the predictability of future operating cash flows. The findings on the effects of conservatism on information quality or asymmetry are mixed and varied. Some studies find that conservatism spurs companies to improve the quality of their accounting information (Fan and Zhang, 2012). Others, however, find that conservatism leads to biases and errors in accounting

reports, thereby reducing reliability of such reports (Nishitani, 2010). However, when the news is extremely negative, unconditional conservatism actually correlates positively with errors in forecast. Gigler and Hemmer (2001) argue that conservative accounting leads to more timely voluntary disclosures while Artiach and Clarkson (2011) comment that the major benefit of conservatism is in its signaling effect, which creates an impression Informativeness of earnings is most critical in contracting and a large number of studies are published in this context (Jackson and Liu, 2010; Betty et al., 2008; Chen et al, 2007) Another strain of thought is that conservatism contributes to information asymmetry (Liu, 2010; LaFond and Watts, 2008) 5 Expressed in another way, the informativeness of earnings is critical for valuation purposes as it provides information on the cash flow of a firm (Balachandran and Mohanram, 2011; Bandyopadhyay et al., 2010; Hui et al, 2009; Chen et al, 2007; Zhang,

2000) 4 5 Source: http://www.doksinet of better quality information, rather than actually improving the quality of accounting information. Recent studies by Sohn (2012), Lara et al. (2011), Bandyopadhyay et al (2010), Jackson and Liu (2010), Kim and Pevzner (2010), Dichev and Tang (2008), LaFond and Watts (2008) amongst others continue to generate conflicting findings on the usefulness of conservatism in financial reporting and it seems that the jury is still deliberating. In addition, there is currently no study conducted to test and compares the robustness of different measurement models. There is also limited effort to address the inconsistency and inconclusiveness of research findings specifically with reference to the use of different conservatism measurement methods and their impact on different industries. A review of the literature on accounting conservatism has revealed two pertinent issues contributing to the inconsistent findings. They are: a. Definitions of accounting

conservatism are vague and arbitrary, and b. There is an absence of a single, comprehensive, and authoritative model for quantification and measurement of accounting conservatism. The next sub‐section will briefly discuss the definitions of conservatism and the major measurement methods used in current research. Authoritative Definition of Conservatism Accounting conservatism can be defined vaguely as ‘exercise of prudence’ or ‘the exercise of caution’ (Givoly and Hayn, 2000). A popular adage describes accounting conservatism as “anticipate no profits, but anticipate all losses” (Watts, 1993). The above definitions, while vague and all encompassing, are not helpful in setting up the theoretical framework for research in accounting conservatism. 6 For instance, Source: http://www.doksinet conservatism in accounting needs to be clearly distinguished from management or business conservatism, which often refers to prudent risk taking behavior for operating or business

decisions. Accounting conservatism is about prudence in the presentation of accounting information and statement of accounting numbers, as for example, recognition and quantification of impairment charges in the income statement due to a prolonged and substantial downward adjustment in the valuation of an asset which has already happened. The economic or trigger event may or in fact may not have happened. For instance, provision for doubtful debts pertains to judgment on the collectability of debts where the uncollectibility of the debt has not happened. The quantum of impairment charge or provision of doubtful debts will invoke different degrees of accounting conservatism as intended by the preparer of financial statements. The GAAP’s definition of conservatism principle is more precise in clarifying prudence to be a choice to state a lower income and net asset whenever the choices exist. This means a lower level of revenue or a higher level of expenses to be recognized whenever

there is a choice or judgement to be made. However, it describes conservatism in a relative sense, in comparing the effects of two or more acceptable accounting choices. It does not provide a solution to the measurement of conservatism as often the benchmark is not defined or is elusive or a matter of judgment. Some scholars fill in the measurement gap by proposing the benchmark for measurement. One of the most recently accepted definitions is “systematic undervaluation of equity relative to economic value” (Watts, 2003a; Givoly et al., 2007) This definition is supported and further clarified by Feltham and Ohlson (1995) to be the downward bias of book value of a firm’s equity as compared to its market value, with market value as the benchmark measure of the true economic value of a firm’s equity. Under this 7 Source: http://www.doksinet definition, any form of earnings report that results in the book value of the entity being lower than its market value would be

considered conservative. The definition of accounting conservatism is also made more complex by adding a time dimension to accounting conservatism. Accounting practices prescribe rules and guidance not only to determine the amount to be recognized but also the timing of recognition. A number of textbook writers incorporated the time dimension into their conservatism definition, mainly stating that accounting conservatism is the slower recognition of income as compared to recognition of expenses (Wolk et al., 1989; Davidson et al., 1985)6 Alternatively, accounting conservatism is the under‐reporting of earnings as compared to cash flows from operations over a defined period of time (Givoly et al., 2007) This carries the notion that there will be convergence between cash flows from operations and net income over a reasonably long period of time. As a result of the complexity and the lack of a conceptually well‐developed and authoritative definition, many researchers have chosen to

provide their own versions of the definition of conservatism to justify different choices of conservatism measures and proxies. Basu defines conservatism as recognition of ‘bad news’ in a more timely manner than ‘good news’ in reported earnings (Basu, 1997).7 Feltham and Ohlson (1995), and Beaver and Ryan (2000) choose to define conservatism as persistently reporting net assets at amounts lower than their market value. Penman and Zhang (2002) focus on the biases in the net operating assets level due to accumulation of layers of reserve (“hidden reserve”) in their definition of accounting conservatism. Currently, there is no single authoritative definition that is applied across accounting This time dimension of conservatism was operationalized by Basu (1997) and henceforth became the most popular measurement of conservatism. 7 Basu’s interpretation implies that conservatism can be represented by the “systematic differences between bad and good news periods in the

timeliness and persistence of earnings”. This is currently the most popular representation of conservatism (Wang et al. 2009) 6 8 Source: http://www.doksinet research. Without such agreement and consensus, comparing findings of different conservatism studies is often like comparing apples with oranges. Measurements of Conservatism An immediate result of the lack of an authoritative definition for accounting conservatism is the proliferation of measurement methods. Various definitions of conservatism emphasize different aspects of conservatism in accounting practices and hence, lead to confusions over the applicability and biases of different methods of measurement of conservatism. One such confusion stems from the lack of a clear distinction and understanding of conditional and unconditional conservatism. Conditional conservatism, as defined by Basu (1997), relates market reactions to a particular good or bad accounting earning or cash flow news. It is event‐based and involves

an external trigger of circumstances, namely, share price returns and thus the term “conditional”. 8 Unconditional conservatism, on the other hand, does not involve such a trigger (Ruch and Taylor, 2011). It is often implicit in the initial recognition of revenues and expenses, and assets and liabilities (Beaver and Ryan, 2005), such as immediate expensing of Research and Development (R&D) costs. These two types of conservatism are largely studied separately, producing a spectrum of conservatism measures that are essentially measuring different purported characteristics of conservative accounting practices. Their distinction has not been well understood until recently where Beaver and Ryan (2005) model the linkage between these two measures through the effect of understating the net assets. However, in recent times, there are numerous criticisms about the validity construct of Basu’s measure and different modifications were made to address the construct issue (Ball et al.,

2011; Patatoukas and Thomas, 2011; Dietrich et al., 2007; Givoly et al, 2007) 8 9 Source: http://www.doksinet Table 1 summarizes the five commonly used measurement models9 in the current state of research (Wang et al., 2009) Basu’s (1997) asymmetric timeliness (AT) conditional conservatism measure is most popular (69%) amongst the 52 papers surveyed by Wang et al (2009) followed by the market‐to‐book (MTB) model (25%). [Table 1 about here] There have been attempts to reconcile unconditional and conditional conservatism measures (Ryan, 2006). Dietrich et al, (2007) find that the AT model is associated positively with the MTB measure over long period and negatively with MTB measure over short period. It is also postulated that conditional conservatism and unconditional conservatism are not completely independent of each other (Ryan, 2006). Therefore, reconciling the two types of conservatism requires a model more complex than a simple summation or subtraction of conditional

and unconditional conservatism measures. The asset‐based model proposed by Beaver and Ryan (2005) to merge these two types of measures is yet to be tested for its effectiveness. The more fundamental questions are: When is conditional conservatism appropriate for use in conservatism impact assessment study as against unconditional conservatism? Are conditional and unconditional conservatism measuring the same underlying accounting conservatism practices of firms? All the above measurement methods are subjected to different criticisms of which the three main criticisms are: a) choice of proxies problem: b) confounding problem and c) incompleteness problem. For the choice of proxies problem, there is a lack of a consistent conceptual framework in the choice of proxies used to construct the 9 See Appendix A for formula and the variables used in computing these five conservatism measures. 10 Source: http://www.doksinet conservatism measure. For example, Basu (1997) uses earnings

while Ball and Shivakumar (2005) use operating cash flow. There is no compelling theory to suggest that both measures are complementary measures of accounting conservatism. Even though earnings and operating cash flow are highly related, these proxies may not measure the same phenomenon which the conservatism measure is trying to capture. The confounding problem is targeted at the use of market share prices or annual stock returns which subject the conservatism measurement to market wide factors or other intervening events which are price sensitive10. These confounding events may have nothing to do with accounting conservatism, for example, the MTB model11. Finally, for the incompleteness problem, this is the classic omitted variable problem where there is no assurance that the relevant factors which measure accounting conservatism are adequately captured by the proxies used in the measurement method.12 For example, in the HR model by Penman and Zhang (2002), impairment losses and

provision for doubtful debts which are huge accounting conservatism playgrounds are omitted from the measurement. Figure 1 provides a diagrammatic summary of the various possible factors which contribute to the conservative accounting practices of a firm which may not be currently captured by all the measurement models. Therefore, it would seem that each measurement has its unique failings and there is currently no consensus as to which measurement model is dominant or conceptually most robust. Due to the different choice of proxies and the above three problems, it is possible that the results in various studies could be driven by the measurement method used. See Manuel and Manuel (2011); Patatoukas and Thomas (2011); Gotti (2008); Roychowdhury and Watts (2007); Givoly et al. (2007) 11 MTB ratio is argued to be biased upward in measuring the level of conservatism in a firm as the effect of economic rent in depressing book value is not recognized and separated from the effects of

conservatism (Roychowdhury and Watts, 2007). 12 For example, Ball etc al. (2011) address this omitted variable problem in Basu’s measure by introducing a fixed‐effect into the regression. 10 11 Source: http://www.doksinet This is not including the possible confounding industry effect where a different industry may have a different degree of accounting conservatism because of its unique accounting practice. For example the degree of conservatism as measured by the HR model which includes R&D will be very different for firms in industries with little or no R&D versus those which are R&D intensive. In response to these limitations, some researchers have chosen to apply more than one measurement model as robustness tests. Wang et al (2009) report that 40% or 21 out of 52 studies reviewed, use more than one measurement model. In addition, the choice of measurement model is also very arbitrary. For example, no paper reviewed justifies the choice of conditional

conservatism measure over unconditional measure. It is also noted that researchers do not use an industry adjusted conservatism measurement model. Instead most would simply use an industry dummy variable in their regression to address the possible industry effects. Considering the differences in the nature of accounting methods for different industries, it is possible that different conservatism measures are suitable for different industries. However, currently, there is no study which tests the applicability of these measurement models to different industries or how to control for the industry effects in a more robust manner. 3 Hypotheses Development This section focuses on developing the hypotheses that conservatism measures are not robust in their applications. The five measures in Table 1 are chosen because they are the most widely applied conservatism measures in published papers on accounting conservatism. The robustness of these measurement models will have implications on

the reliability of the recent conservatism impact assessment studies. 12 Source: http://www.doksinet Out of the five conservatism measures, AT and ACCF models measure the level of conditional conservatism, whereas HR, MTB, and NA models measure the level of unconditional conservatism in firms. Since the relationship between conditional and unconditional conservatism is still under investigation, comparisons between these two groups of conservatism measures will have to be treated with care. Industry Confounding Effects Different industries may have different unique accounting practices which render inter‐industry comparison of conservatism measurements difficult. Conceptually, different industries have unique characteristics that are likely to influence the type and extent of conservatism in their accounting practices. In terms of conditional conservatism, different industries could be measured unfairly as they are sensitive to different kinds of news that impact the

firms over different lengths of investment horizon. For example, oil and gas firms are sensitive to general economic news that often impacts the earnings performance of oil and gas firms over long term, while high technology consumer goods are more sensitive to firm specific news that are more likely to have short term impacts on the firms’ earnings. Hence, it is important to verify whether industry effects have significant implications for the different conservatism measurement models.13 The difference in the duration of investment and earnings persistence is likely to affect the extent of the incremental timeliness of recognizing ‘bad news’ over ‘good news’ under the conditional conservatism measures. Different unconditional conservatism measures may be appropriate for one industry but not for others because However, a majority of empirical conservatism impact assessment research applies conservatism measures arbitrarily to cross‐industry data. Chandra (2011) focusing

on the technology sector is an exception. A general control for industry effect is to use industry dummy variables in the regression analysis. 13 13 Source: http://www.doksinet of the relative importance of their income statement and balance sheet items. For example, firms from high technology industry which are light on assets because of immediate expensing of R&D costs and heavy importance of human resource capital could be measured wrongly when compared with capital intensive industries like utilities that have a significant amount of long term assets. Different conservatism measures will capture different aspects of high technology firms. Hence, it is postulated that: H1: Different conservatism measurement models will rank different industries differently. Impact Assessment Study Using Different Conservatism Measures Measurement of conservatism is an integral and important part of conservatism impact assessment studies. Empirical research in this area often regress

conservatism score against proxies for earnings quality and hence different conservatism scores may impact the results of such regressions differently. Only 15% of the published studies (8 of 52) used three or more conservatism measures for robustness testing and 60% of the studies (31 of 52) uses only one measurement model (Wang, et al., 2009) If different measurement models produce different results, the reliability of studies using one measurement model will become questionable. We use Penman and Zhang’s (2002) study on the relationship between accounting conservatism and earnings persistence as the benchmark conservatism impact study. Our study does not examine the soundness of the proposed theoretical framework used by Penman and Zhang. Rather, it investigates whether the same conclusion between accounting conservatism and earnings persistency as documented by Penman and Zhang (2002) can be established using different unconditional 14 Source: http://www.doksinet conservatism

measurement models. We replicate Penman and Zhang (2002) and substitute their unconditional conservatism Q‐Score with the MTB and NA unconditional conservatism scores.14 The robustness of the different conservatism scores will be tested. The following states the null hypothesis for our study: H2: Different unconditional conservatism measures will produce similar results as found in Penman and Zhang’s (2002) earnings persistence study. 4. Data and Results Data Data for this study are taken from COMPUSTAT Annual Industrial and Research files including non‐survivors and holding period return from CRSP monthly stock/security files from the period 1975 to 2010. The test of hypothesis H2 requires the replication of Penman and Zhang’s study. To be consistent with their research methodology, a comparative pool of samples from 1975 to 1997 similar to Penman and Zhang’s is used. This is augmented by additional firm‐years taken from 1998 to 2010 The initial query on COMPUSTAT

yielded 96,085 firm‐years across all industries from 1975 to 2010. The 96,085 firm‐years were then sorted according to their 2‐digit SIC, and the top ten industries with the most number of available firm‐years were chosen for analysis. Various wholesale and retail trading companies under the 2‐digit SIC 50 to 59 were combined to form a single wholesale and retail industry. We justified this grouping by the similarities in their accounts such as near zero R&D and high inventory turnover. This resulted in a total of 65,786 firm‐years (685% of the initial sample) from 1975 to 2010. Further filtering required removal of firm‐years where any We did not include the conditional score in the analysis because of the conceptual difference between the conditional and unconditional conservatism measures since Wang et al. (2009) document significant separation between these two types of conservatism measures. 14 15 Source: http://www.doksinet of the five conservatism scores

could not be computed meaningfully. The final sample consisted of 43,434 firm‐years for ten industries. The filtering process, the ten selected industries, their SIC and the number of firm‐years are found in Table 2. Each firm year‐ data contains 27 different variables used to compute the five conservatism scores and core‐RNOA. The variables and the formula for computing the five conservatism scores are found in Appendix A. [Table 2 about here] While the original sample includes firm‐years from 1975 to 2010, the final sample consists only of firm‐years from 1988 onwards because of the extensive data requirements for the construction of the conservatism scores. Out of the 43,434 firm‐ years in the final sample, 19,931 firm‐years (45.9% of the sample) are between 1988 and 1997, replicating the data period Penman and Zhang used in their study15 and the remaining 23,503 firm‐years (54.1% of the sample) are from 1998 and 2010 Results Following Penman and Zhang’s (2002)

data approach, firm‐year data were taken from 1975‐1997. An extended period of data from 1988‐2010 was analyzed to further test the robustness of the test results across a longer period of time, as well as to provide an update to the results of Penman and Zhang (2002) by including more recent firms’ information. Table 3 summarizes the descriptive statistics of nine selected variables16 for the two time periods of analysis, 1988‐1997 (10 years) and 1988‐2010 (23 years). Penman and Zhang (2002) use a sample of 29,796 firm‐years for their final regression. There are 27 variables used in this study. For brevity in reporting, these nine key variables are selected The descriptive statistics of the 27 variables are available upon request. 15 16 16 Source: http://www.doksinet [Table 3 about here] From Table 3, it is evident that firm‐years from both periods of analysis show significant positive skewness in most of the major accounts. This means that data representation is

biased towards larger firms. The degree of skewness remains fairly consistent across both periods of analysis, showing that their compositions of firms of different sizes are likely to be consistent. However, the maximum value of each selected account for the period 1988‐2010 is much higher than the maximum value of the similar accounts for the period 1988‐1997. This led to a significant difference in the means of the key variables between 1988‐1997 and 1988‐2010. Market value, total asset (AT), and expense accounts (DP, XAD, and XRD) nearly doubled in the extended period showing that firms have grown significantly in size and asset base subsequent to 1997. This has a significant implication for accounting conservatism as growing firms often accumulate hidden reserves when increased assets are expensed instead of being capitalized and amortized over future years (Penman and Zhang, 2002). LIFO reserve remains relatively similar (6.600 in the earlier period versus 7051 in the

final sample) probably due to the popularity of FIFO inventory accounting in recent years, as outside the US, LIFO is often not allowed. On the basis of the variations of accounts over time, asset based conservatism measures, especially HR, are likely to observe an increase in conservatism since the growth in firm assets is likely to increase the size of hidden reserves. Therefore the changes in level of conservatism over the years may not be measured consistently across different measurement models. Hence the consistency of conservatism measurement models when applied to different industries was further analyzed over an extended period of 1988 to 2010. 17 Source: http://www.doksinet Table 4 provides the breakdown of selected accounts of firms in Table 3 according to industry17. For each industry, each account mean is compared to the full sample mean and presented as a percentage. Each industry has clearly different levels of accounts. For example, Oil & Gas (13), Business

Services (73) have typically low inventory balances (almost 20% of overall sample average) while Transportation Equipment (37) and Trade (50‐59) have on average very high balances (295% and 163% of overall sample mean respectively). On the other hand, Trade (50－59) has near zero R&D expense (1% of overall sample mean) while Transportation Equipment (37) incurs high R&D expense (273% of overall sample mean). Different measurement models consisting of different combinations of accounts are hence likely to measure conservatism of these companies differently. In order to test the significance of the differences in accounts between industries, ANOVA was performed to test if inter‐ industry variance in each variable is significantly higher than intra‐industry variance in the variables. The results show that for all 9 variables in Table 4, inter‐industry variance is significantly higher than intra‐industry variance. [Table 4 about here] The results show a strong support

for the grouping of firms according to industries, as the variables vary significantly from industry to industry. The conservatism measurements associated with these accounts are also likely to vary from industry to industry. As no single conservatism measure includes all relevant variables, their suitability and comprehensiveness in measuring a firm’s level of accounting conservatism is likely to vary from industry to industry. 17 Due to limitation of space, the numbers for 1988‐1997 are not presented but are available on request. 18 Source: http://www.doksinet Results of the Robustness of Conservatism Scores at Industry Level To investigate H1, accounting conservatism measurement models were tested for their consistency in their application to different industries. Conditional conservatism scores, asymmetric timeliness (AT) and asymmetric accruals to cash flow (AACF) were computed according to the regression model described in Appendix A. Unconditional conservatism

scores, including market‐to‐book ratio of equity (MTB), hidden reserves (HR), and negative accruals (NA), were computed for each firm‐year. The regression results for the AT and AACF models are summarized in Table 5. Table 5 shows significant difference between timeliness of recognition of good news and bad news for a majority of industries. Regression coefficients obtained under the AT model (ranges from 0.109 to 0700) show less variation between industries as compared to the AACF model (ranges from ‐0.985 to 19688) In fact, the results of the AT model are relatively stable between industries and across time18. For the AACF model, for the period 1988‐1997 (Panel C), four out of ten industries’ (SIC 20, 28, 38, 48) β1 is not significant at 10% level of significance. This weakens the model’s applicability on the industry level. In addition, for the same period, two out of the ten industries (SIC 37 and 50‐59) are significantly negative, suggesting that firms recognize

bad news less timely as compared to good news, which is inconsistent with the results obtained by the AT model. The results in Panel D for the period of 1988‐2010 for the AACF model are not much better. This shows that between conditional conservatism models, the measurement results do not always agree. [Table 5 about here] 18 The exception is SIC 48 for the period 1988‐2000 which shows insignificant results. 19 Source: http://www.doksinet In order to investigate the extent of inconsistency between conservatism measurement models in their application to different industries, the industries were ranked based on the score obtained using each conservatism measurement model. Raw conservatism scores and ranking of industries for both 1988‐1997 and 1988‐2010 are summarized in Table 6. The scores and ranks were further analyzed to investigate if there was any agreement between conservatism models in their measurement of industry conservatism. [Table 6 about here] Table 6 shows

that the rankings of conservatism measures for the 10 industries are very different. For both periods, there is no single industry that is ranked consistently across all five conservatism measures. Some of the differences in ranking can be stark. For example, in Panel A, Communications (48) is ranked as the second least conservative by AT score but it is ranked as the most conservative by NA and MTB scores. From the descriptive statistics (Table 5), it can be seen that Communications (48) has extremely high accruals, which is 834% of full sample mean, and high depreciation, which is 788% of full sample mean. Hence its accrual balance is likely to cause an upward bias in its conservatism score measure using negative accruals only. Evidently, a high depreciation level indicates a high likelihood of a deflated book value through accelerated depreciation, and hence leads to an upward biased MTB conservatism score. The most consistently ranked industry is the Oil & Gas (13) as it is

ranked as most conservative by all three measures: AT score, HR score and NA score. Panel B provides an even more confusing ranking as only Chemical (28) is ranked by HR and MTB as the most conservative. The average rank of the industries ranged from 26 to 7.6 (24 to 82) in Panel A (B) The standard deviations for Panel A (B) ranged from 20 Source: http://www.doksinet 1.3 to 36 (08 to 40) Figure 2 provides a pictorial summary of the ranking of the ten industries according to each of the five conservatism scores. There is no consistency in the patterns of ranking for these ten industries in Figure 2. [Figure 2 about here] Overall, there is significant disagreement in conservatism score and ranking among the five conservatism measurement models. Ideally, when all conservatism measures are consistent, Figure 2 should show horizontal lines that are parallel to each other. 19 This means that if a particular industry has high or low accounting conservatism, its rank would be consistently

high or low across all five measurement models. However this is not the case Therefore, there is strong support for H1 Both conditional measures and unconditional measures rank the industries very differently. There is no clear relationship identified amongst the conditional or unconditional conservatism measures. The next section presents the results of classification tests Classification Overlap for Unconditional and Conditional Measures To conduct classification tests, each industry is classified under high conservatism (top 3 in conservatism score) or low conservatism (bottom 3 in conservatism score) according to each conservatism model. If more than one conservatism model classifies a particular industry in the same category, high or low, this is considered an overlap in classification. The greater the overlap, the greater the agreement amongst different conservatism models. Figures 3 and 4 present the results The same inconsistencies are observed when we compared the

conditional conservatism scores AT with AACF or the unconditional conservatism scores (HR, NA and MTB) separately. Figures are available on request from authors. 19 21 Source: http://www.doksinet of classification for the periods 1988‐1997 and 1988‐2010 for the conditional and unconditional conservatism measures respectively. [Figures 3 and 4 about here] From Figure 3 Panel A, for the period 1988‐1997, for high conservatism, two of the three industries (SIC 35 and 36) are classified similarly. However for the low conservatism classification, only one industry (SIC 50) is classified as such. For Panel B, for the period 1988‐2010, only one industry is classified consistently for both high (SIC 73) and low (SIC 50) conservatism categories. The AT model of classification for high conservatism is totally different between the two periods, namely, SIC 20, 35 and 36 for 1988‐1997, and SIC 13, 37 and 73 for the period 1988‐2010. However the low conservatism classification is

relatively stable for both periods with the same industry classified as low conservatism in both measures in both time periods (SIC 50). Therefore, at best, the conditional conservatism measures of AT and AACF give a similar classification for two of the three industries (67% classification consistency), and at worst one in three industries (33%). From Figure 4 Panel A, for the period of 1988‐1997, there is no consistent ranking by all the three unconditional conservatism measures for any industry as a high conservatism industry. In fact, at best, for any two unconditional measures, only one industry is classified similarly. For example, using NA and MTB, only Communications (48) is classified similarly as high conservatism. For the low conservatism classification, only the Oil & Gas (13) is classified as low conservatism in all three unconditional measures. The same is observed for the period 1988 to 2010 However, a unique difference is that for both MTB and HR measures, the

three industries are consistently 22 Source: http://www.doksinet classified as low conservatism (SIC 13, 37 and 50). The Venn diagrams in both Panels A and B suggest that consistency in classification is limited except for the low conservatism for the Oil & Gas industry and between MTB and HR measures. The classification overlaps amongst all five conservatism measurement models are also examined and the results are summarized in Table 7. There is no unanimous agreement on the classification of any industry for all five measures, namely, the column with “5 overlaps” is a null set. In Panels A and B, the classification of Oil and Gas (13), and Trade (50) is comparatively the most consistent for the period 1988‐1997 as low conservatism by four of the five measures. In general, most industries are classified similarly only by two of the five measures. Panels C & D provide an almost similar conclusion except where four measures classify Business Services (73) as high

conservatism and Trade (50) as low conservatism similarly. [Table 7 about here] Apart from classification test conducted at industry level, classification test was also performed at firm‐year level for unconditional conservatism measures for which firm‐specific scores are available. The rationale for conducting the additional firm‐year analysis is to understand the degree of overlap in the classification of individual firm‐ years. Figure 5 summarizes the results obtained for the classification overlap of all firm‐years in the final sample. From Panels A and B, for the period 1988‐1997, for high (low) conservatism, only 9.8% (127%) of the sample is similarly classified by the three measures. From Panels C and D, for the period 1988‐2010, for high (low) conservatism, only 10.2% (119%) of the sample is similarly classified The results suggest that less 23 Source: http://www.doksinet than 13% of the firm‐years are classified similarly by these three measures as high

or low conservatism. [Figure 5 about here] Table 8 summarizes the classification overlap of firm‐years in each industry. Oil & Gas (13) has the most consistent classification for high and low conservatism for both periods, ranging from 24% to 29%. For the rest of the industries, the consistency of classification ranges from a low of 4% for Communications (48) for high conservatism for the period 1988‐1997 and a maximum of 15% for Chemicals (28), Electronics (36), Transportation Equipment (37) and Communications (48) for low conservatism. The level of consistency is very low as the overall consistency classification ranges from a minimum of 1.2% (4% * 30%) to a maximum of merely 8.7% (29% * 30%) of the total firm‐years. [Table 8 about here] Overall, the classification overlap among the three unconditional conservatism measures is consistently low at below 10% for both high and low conservatism. This suggests that when conservatism measures are applied to firm‐year data

across industries, only one out of every ten firm‐years investigated is measured and classified consistently across three unconditional conservatism measures. This finding severely weakens the reliability of using only one conservatism measure to determine a firm’s level of unconditional conservatism. However, using more than one conservatism measure may lead to conflicting results. In Penman and Zhang (2002), firms are classified into high conservatism, moderate conservatism and low conservatism based on the HR score. The portfolios of firms classified under each category are likely to be 24 Source: http://www.doksinet different if a different model is chosen to compute the conservatism measure. Will the use of a different conservatism measure result in a different conclusion for their study? This question is addressed in the next sub‐section. Analysis of Results of Robustness Test Based on Penman and Zhang’s Regression In the current study, there was sufficient data to

calculate Q‐Scores for 36,839 firm‐years for the period 1988‐2009 (22 years). The frequency curve of core‐RNOA and Q‐Score variables showed that there were a few but very extreme value outliers on both ends. Extreme value outliers were removed to obtain reliable regression results Since the frequency curves of variables displayed very thin tails, data outside one standard deviation for each variable were removed. The filtered sample contained 33,943 firm‐ years, which was 92% of the final sample. Table 9 summarizes the key statistics of Q HR score of the filtered sample, and these values are presented next to the Q‐Score statistics computed by Penman and Zhang. Although data were retrieved from the same source (CRSP and COMPUSTAT), and computation of Q‐Score in this study followed the methodology described by Penman and Zhang, differences in data were expected due to the difference in the data periods and the extensive data requirements in constructing the rest of the

conservatism measures. [Table 9 about here] Earnings persistence regressions were performed using Q HR according to the equation used by Penman and Zhang as the control for data variation. If the data used in this study were not significantly different from that used by Penman and Zhang, it was expected that the same results as the regressions performed by Penman and Zhang 25 Source: http://www.doksinet would be produced. To replicate Penman and Zhang’s study, 22 annual cross sectional regressions were performed using computed Q HR for the filtered sample. Coefficient for Q HR was obtained for each year. One‐sample t‐test was then performed to test if the mean of the 22 coefficients was significantly different from zero. As shown in Table 10, the results of the t‐test for Q HR coefficients in this study are significant at 1%, as were those obtained by Penman and Zhang. [Insert table 10 about here] The significance of Q HR coefficient obtained in this study also suggests

that Penman and Zhang’s model is robust across periods and still applies to a more recent time period although the magnitude of the coefficients is smaller in our study. However, when the same tests were performed for Q MTB and Q NA, the results obtained were not significant. Hence, further regressions were performed for each industry to further investigate the relationship between different Q‐Scores and next year Core‐RNOA. Table 11 summarizes the results of the three regressions for each Q‐Score by running a single pooled regression for all firm‐years for the sample. The results show that coefficients of Q HR and Q MTB are significant at 1%. Q NA model shows a slightly weaker result but still significant at 5%. The overall results produced by different Q‐ Scores are fairly consistent. However, we next examined the regressions performed at industry level. [Insert Table 11 about here] Stacked regressions were performed for each of the Q‐Scores. Dummy variables were

assigned to industries for nine out of the ten industries. Business (73) was the 26 Source: http://www.doksinet base industry and its regression coefficient was subsumed in the constant term. The Q‐ Score coefficients were obtained for each industry and summarized in Table 12. The results were very inconsistent across industries. For Q HR, although its coefficient in earnings persistence regression was shown to be significant in both annual cross‐ sectional regression and single pooled regression, it was not significant for three out of nine industries (37, 38 and 48). In addition, two of the nine industries (13 and 50) had significantly negative coefficients. For Q MTB, only two out of nine industries (20 and 50) coefficients were significant. Finally, for Q NA, all coefficients were not significant even at 5% level of significance. Such inconsistent results raise two issues First, using different Q‐Scores leads to different conclusions about the relationship between

conservatism and one‐year‐ahead RNOA. Adoption of Q HR score leads to a positive conclusion while adoption of Q NA score leads to a negative conclusion. The results obtained using Q MTB are mixed. Secondly, the regression results are not consistent across industries. This means that the ability of Q‐score to improve the predictability of next year Core‐RNOA may vary from industry to industry. The conservatism measures are not robust in their application to different industries and to Penman and Zhang’s earnings persistence regression. The results obtained in this study show that conditional and unconditional conservatism measures score and rank different industries very inconsistently and hence provide strong support for H1. For the second part of the study, although results from the single pooled regression produce significant results for all three Q‐Scores, results obtained using annual cross‐sectional regressions and industry stacked regressions show that there are

significant inconsistencies in the results. Hence, on the basis of the findings, this study rejects hypothesis H2. 27 Source: http://www.doksinet 5. Conclusion As discussed, there are three main problems with current research on conservatism. First, definitions of accounting conservatism are vague and arbitrarily interpreted. Secondly, there is no authoritative model for quantification and measurement of accounting conservatism. Thirdly, measurement models are chosen and applied arbitrarily in empirical research to assess the impact of conservatism. The above‐mentioned issues have led to significant disagreements in current research findings. Our study seeks to document the potential inconsistencies among different conservatism measurement models and their impact on conservatism impact studies. This study hence conducted two tests to investigate the robustness of five most influential conservatism measurement models. The first tested the robustness of these measurement models

in their application to different industries. The second tested the ability of different unconditional conservatism measurement models to produce consistent results in Penman and Zhang’s earnings persistence regression. The results suggest that both conditional conservatism measurement models and unconditional conservatism measurement models rank the industries inconsistently. This implies that conservatism measurements at industry level based on a single model are highly unreliable, and comparisons between different industries measured using different models cannot be reliably made. The results of the robustness test based on Penman and Zhang’s earnings persistence regression show that results obtained using Q‐Score computed from different unconditional conservatism measures produce very inconsistent results. This suggests that different unconditional conservatism measurement models may be measuring different conservatism phenomena. Therefore, the results produced by

conservatism impact assessment studies are shown to be dependent on the choice of 28 Source: http://www.doksinet conservatism measure. The results obtained from industry‐based regressions are significantly different from industry to industry, too. This implies that firms in different industries are not homogenous in their earnings persistence response to conservatism. The relationship established between accounting conservatism and predictability of core RNOA in Penman and Zhang’s study may not apply to all industries. Since this study has shown that the choice of conservatism measure can affect the results of the impact assessment study of PENMAN AND ZHANG significantly, rampant disagreements in current research literature are likely a result of inconsistent application of conservatism measurement models. Since 94% of current research adopts one or more of the five conservatism measurement models assessed in this study, and a majority of studies (60%) rely on a single

measurement model, there is sufficient evidence to raise concerns on the reliability and generalizability of results obtained from these studies. These results may not be replicated using different conservatism measures. These results also have limited applicability on the industry level due to heterogeneous earnings response to conservatism among industries. Areas of Interest for Future Research It has been established that studies on accounting conservatism produce unreliable results if they rely only on one or two current conservatism measurement models, especially when applied to cross‐industry samples. Hence, future research should be conducted to explore, investigate, and propose ways to improve current conservatism measurement methodology to ensure that reliable results can be obtained. There are several interesting areas that scholars may look into First, there is a need to refine the concept and definition of conservatism and to propose a measurement model that is

consistent with the concept and definition of conservatism 29 Source: http://www.doksinet so articulated. This is potentially a project of immense scale, but if successful, it will result in an authoritative measurement for conservatism. This authoritative conservatism measure will also have greater potential in practical application to improve regulation policies and management decision‐making. Secondly, since industry effects are documented in the conservatism measures in this study, industry specific conservatism measures should be derived and research may need to be conducted on industry specific impact assessment of accounting conservatism. produce more reliable and more specific results. 30 This is likely to Source: http://www.doksinet Reference Artiach, T. and P Clarkson 2011 Conservatism, disclosure and the cost of equity capital Accounting and Finance 51(1): 2‐49. Balachandran, S. and P Mohanram 2011 Is the decline in the value relevance of accounting driven by

increased conservatism? Review of Accounting Studies 16:272‐301. Ball, R. and L Shivakumar 2005 Earnings quality in UK private firms: comparative loss recognition timeliness. Journal of Accounting and Economics 39(1): 83‐128 Bandyopadhyay, S., C Chen, A Huang and R Jha 2010 Accounting conservatism and the temporal trends in current earnings’ ability to predict future cash flows versus future earnings: evidence on the trade‐off between relevance and reliability. Contemporary Accounting Research 27(2): 413‐460. Basu, S. 1997 The conservatism principle and the asymmetric timeliness of earnings Journal of Accounting and Economics 24(1): 3‐37. Beatty, A., Weber, J, and JJ Yu 2008 Conservatism and Debt Journal of Accounting and Economics 45(2‐3): 154‐174. Beaver, W.H and SG Ryan 2000 Biases and lags in book value and their effects on the ability of the Book‐to‐Market ratio to predict book return on equity. Journal of Accounting Research 38(1): 127‐148. Beaver, W.H and

SG Ryan 2005 Conditional and unconditional conservatism: Concepts and modeling. Review of Accounting Studies 10(2‐3): 269‐309 Chandra, U. 2011 Income conservatism in the US technology sector Accounting Horizons 25(2): 285‐314. Chen, Q., T Hemmer and Y Zhang 2007 On the relation between conservatism in accounting standards and incentives for earnings management. Journal of Accounting Research 45(3) (June): 541‐565. Davidson, S., Stickney, CP, and RL Weil 1985 Intermediate Accounting: Concepts Methods and Uses, 4th Edition. USA: Dryden Press Dichev, I. and V Tang 2008 Matching and the changing properties of accounting earnings over the last 40 years. The Accounting Review 83(6): 1425‐1460 Dietrich, J.R, KA Muller and EJ Riedl 2007 Asymmetric timeliness tests of accounting conservatism. Review of Accounting Studies 9:495‐521 Fan, Q. and Zhang XJ 2012 Accounting conservatism, aggregation, and information quality. Contemporary Accounting Research: 29(1):38‐56 31 Source:

http://www.doksinet Financial Accounting Standards Board. Conceptual framework for financial reporting – Chapter 1, The objective of general purpose financial reporting, and Chapter 3, Qualitative characteristics of useful financial information. Statement of Financial Accounting Concepts No. 8, September 2010 Feltham, G.A, and JA Ohlson 1995 Valuation and clean surplus accounting for operating and financial activities. Contemporary Accounting Research 11(2): 689‐ 731. Gigler, F.B and T Hemmer 2001 Conservatism, optimal disclosure policy, and the timeliness of financial reports. The Accounting Review 76(4): 471‐493 Gotti, G. 2008 Conditional conservatism in accounting: new measure and tests of determinants. Conference Paper: American Accounting Association (AAA) 2008 Annual Meeting Proceedings Series. Givoly, D. and C Hayn 2000 The changing time‐series properties of earnings, cash flows and accruals: Has financial reporting become more conservative? Journal of Accounting &

Economics 29(3): 287‐320. Givoly, D., C Hayn and A Natarajan 2007 Measuring reporting conservatism Accounting Review 82(1): 65‐106. Hui, K.W, Matsunaga, S and D Morse 2009 The impact of conservatism on management earnings forecasts. The Journal of Accounting and Economics 47(3): 192‐207. Jackson, S. and X Liu 2010 The allowance for uncollectible accounts, conservatism, and earnings management. Journal of Accounting Research 48(3): 565‐601 Kim, M. and W Kross 2005 The ability of earnings to predict future operating cash flows has been increasing – not decreasing. Journal of Accounting Research 43(5): 753‐780. Kim, B. H and M Pevzner 2010 Conditional accounting conservatism and future negative surprises: an empirical investigation. Journal of Accounting and Public Policy 29(4): 311‐329. LaFond, R. and RL Watts 2008 The information role of conservatism The Accounting Review 83(2): 447‐478. Lara, J.MG, BG Osma and F Penalva 2011 Information effects of conservatism in

accounting. Working Paper: Universidad Carlos III de Madrid, Universidad Autonoma de Madrid, and University of Navarra ‐ IESE Business School. Li, J. 2010 Accounting conservatism, information uncertainty and analysts’ forecasts Working Paper: Carnegie Mellon University. 32 Source: http://www.doksinet Manuel, C., and NN Manuel 2011 Consistent estimation of conditional conservatism Working Paper: University of Jaen and Universidad Carlos III de Madrid. Nishitani, J. 2010 Type I and type II errors of conservatism Conference Paper: Canadian Academic Accounting Association (CAAA) Annual Conference Proceedings Series. Patatoukas, P.N, and JK Thomas 2011 More evidence of bias in the differential timeliness measure of conditional conservatism. The Accounting Review 86(5): 1765‐1793. Penman, S. and X Zhang 2002 Accounting conservatism, the quality of earnings, and stock returns. The Accounting Review 77(2): 237‐264 Roychowdhury, S. and RL Watts 2007 Asymmetric timeliness of

earnings, market‐to‐ book and conservatism in financial reporting. Journal of Accounting and Economics 44(2): 2‐31. Ruch, G.W, and G Taylor 2011 Accounting conservatism and its effects on financial reporting quality: a review of the literature. Working Paper, University of Alabama Ryan, S.G 1995 A model of accrual measurement with implications for the evolution of the book‐to‐market ratio. Journal of Accounting Research 33(1): 95‐112 Ryan, S.G 2006 Identifying conditional conservatism European Accounting Review 15(4):511‐525. Sohn, B.C 2012 Analyst forecast, accounting conservatism and the related valuation implications. Accounting and Finance: 52(1):311‐341 Sterling, R. 1970 Theory of the measurement of enterprise income USA: Accounting Publication of Scholars Book Co. Wang, R.Z, CO Hogartaigh and TV Zijl 2009 Measures of accounting conservatism: a construct validity perspective. Journal of Accounting Literature 28: 165‐203 Watts, R.L 1993 A proposal for research on

conservatism Conference Paper: American Accounting Association (AAA) Convention, San Francisco, August 1993. Watts, R.L 2003a Conservatism in accounting part I: Explanations and implications Accounting Horizons 17(3): 207‐222. Watts, R.L 2003b Conservatism in accounting part II: Evidence and research opportunities. Accounting Horizons 17(4): 287‐301 Wolk, H. I, Francis, JR and MG Tearny 1989 Accounting Theory: A Conceptual and Institutional Approach. USA: PWS‐KENT Publishing Company Zhang, X.J 2000 Conservative accounting and equity valuation Journal of Accounting & Economics 29(1): 125‐149. 33 Source: http://www.doksinet Appendix A: Summary of Formulas and Computation Variables This appendix presents a summary of the formulas used for the computation of the conservatism measures and regressions used in this study. Conservatism Measures 1 Formula Data variables needed Basic earnings per share Fiscal year closing price Monthly Return AT‐Score ∗ 2 AACF‐Score

Rit is calculated by compounded CRSP monthly return data is a dummy variable where it is 1 if the return is negative. AT‐Score = ∗ CFO = Earnings before extraordinary items ‐ accruals Accruals = Δ inventory + Δ debtors + Δ other current assets – Δ creditors – Δ other current liabilities – depreciation is a dummy variable where it is 1 if the CFO is negative. AACF‐Score = 3 HR‐Score = reported LIFO reserve for firm i in fiscal year t = Capitalized R&D expense – amortization = Capitalized advertising expense –amortization = Total assets – total liabilities – preference equity (book) HR‐Score = 4 NA‐Score ∆ ∆ ∆ . ∆ ∆ ^NA‐Score = (‐1)(NOACC) 34 Earnings before extraordinary items Inventory Debtors Other current assets Creditors Other current liabilities Depreciation LIFO reserve R&D expense Advertising expense Total assets Total liabilities Preference Equity Net Income Depreciation Cash flow from Operations Account

Receivable Inventory Prepaid Expense Account Payable Tax Payable COMPUSTAT/ CRSP Code EPSPI PRCC F RET IB INVT RECT ACO AP DLC DP LIFR XRD XAD AT LT PSTK NI DP OANCF RECT INVT XPP AP TXP Source: http://www.doksinet 5 ∗ MTB‐Score Number of outstanding shares at fiscal year end Fiscal year closing price Total asset Total Liabilities Preference equity ^ MTB‐Score = CSHO PRCC F AT LT PSTK Robustness Test of Conservatism Measures using Penman and Zhang’s (2002) Methodology Item 1 Formula Data Variables ∗ 1 Core‐RNOA ^Core Operating Income = Operating income after depreciation + Interest Expense ^ 2 Q‐Score 0.5 ^ 3 = Total assets – total liabilities – preference equity (book) 0.5 = Conservatism Score for firm i in fiscal year t Regression 35 Operating income after depreciation Interest expense Total assets Total liabilities Preference equity (book) Compustat/ CRSP Code OIADP XINT AT LT PSTK Source: http://www.doksinet Table 1: Five Commonly Used

Measurement Methods This table presents the frequency of the five conservatism measurement methods surveyed by Wang et al (2009). S/N Measurement Method Contributor Frequency of Use (Wang et al., 2009) Type of Conservatism Measures 1 Asymmetric Timeliness (AT) Model Basu (1997) 36 of 52 papers (69%) Conditional Measure 2 Asymmetric Accrual to Cash‐Flow (AACF) Model Ball and Shivakumar (2005) 7 of 52 papers (13%) Conditional Measure 3 Hidden Reserve (HR) Model 9 of 52 papers (17%) Unconditional Measure 4 Negative Accruals Measure (NAM) Model Givoly and Hayn (2000) 10 of 52 papers (19%) Unconditional Measure Market‐to‐Book (MTB) Model Feltham and Ohlson (1995) Ryan (1995) Beaver and Ryan (2000) 13 of 52 papers (25%) Unconditional Measure 5 Penman and Zhang (2002) 36 Source: http://www.doksinet Table 2: Sample Filtering Steps This table summarizes the sampling criteria that result in the final sample of 43,434 firm years over 23‐year period of

1988‐2010. Figures in parenthesis are percentages relative to the ten‐industry sample. The final sample consists of firm‐years from 1988 to 2010 because of the data demands in computing the five conservatism scores even though the original period of study was from 1975 to 2010. Panel A : Derivation of the Final Sample Sample Size (%) 65,786 (100.0%) 3,142 (4.8%) 2,458 (3.7%) 60,186 (91.5%) 10,114 (15.4%) 6,638 (10.1%) 43,434 (66.0%) 19,931 (45.9%) 23,503 (54.1%) 43,434 (66.0%) Description Firm‐years in the ten‐industry sample Less: firm‐years with negative or zero book value (BV) Less: firm‐years with negative or zero net operating assets (NOA) Firm‐years before screening for unavailable conservatism score Less: firm‐years without NA‐Score Less: firm‐years without returns data Firm‐years with complete information for all five conservatism measures (Final Sample) Firm‐years from 1988 to 1997 Firm‐years from 1998 to 2010 Final Sample covering 1988 to 2010

Panel B: Distribution of the Final Sample Amongst the Ten Industries Name of Industry Oil and Gas Food and Kindred Products Chemicals and Allied Products Industrial and Commercial Machinery and Computer Equipment Electronics and other Electrical Equipment Transportation Equipment Instruments and Related Products Communications Retail and Wholesale Trade Business Services Final Sample 37 2‐digit SIC 13 20 28 35 36 37 38 48 50‐59 73 Firm‐ years 2,139 1,916 5,752 4,672 5,910 1,531 4,840 2,081 8,392 6,021 43,434 Source: http://www.doksinet Table 3: Descriptive Statistics for the Final Sample This table presents the descriptive statistics of the key variables for the two periods of analysis adopted in this study, 1988‐1997 and 1988‐2010. MKT denotes market capitalization of the firm, AT denotes total asset, INVT denotes inventory, LIFR denotes LIFO reserve, EPSPI denotes basic earnings per share, ACC denotes accruals, DP denotes depreciation, XAD denotes advertising expense,

and XRD denotes R&D expense. Panel A ‐ 1988‐1997 Descriptive Statistics MKT 19931 AT 19931 INVT 19931 LIFR 19439 EPSPI 19931 ACC 19931 DP 19931 XAD 6455 XRD 13961 25 1224.190 81.152 5649.324 11.765 0.062 164758.840 19.915 1042.223 76.007 4712.890 11.103 0.218 132864.000 20.401 137.725 8.584 633.085 12.496 0.000 17665.831 1.109 6.600 0.000 51.276 25.118 ‐88.000 2123.000 0.000 0.416 0.300 1.753 ‐0.646 ‐39.570 28.020 ‐0.180 ‐48.242 ‐1.211 385.251 ‐23.577 ‐18472.000 3979.000 ‐10.323 54.653 3.077 327.618 22.308 ‐4.385 17287.000 0.727 57.287 2.459 216.341 7.147 0.000 3468.000 0.410 50.524 2.397 262.473 10.163 0.000 5522.258 0.231 75 381.279 326.612 50.672 0.000 1.050 0.825 13.814 14.667 12.651 Market Value 43434 AT 43434 INVT 43434 LIFR 42394 EPSPI 43434 ACC 43434 DP 43434 XAD 15370 XRD 31162 25 2935.599 148.387 14141.116 11.690 0.062 467092.880 32.427 2342.280 141.617 11484.951 12.573 0.218 324939.000 33.135 215.505 12.390

918.001 13.259 0.000 35180.000 1.261 7.051 0.000 59.733 24.327 ‐196.100 3003.000 0.000 0.432 0.330 2.024 ‐0.689 ‐52.840 45.510 ‐.240 ‐116.250 ‐2.911 862.805 ‐19.078 ‐47110.156 9417.000 ‐24.465 119.877 5.653 771.365 18.378 ‐4.385 33750.967 1.215 92.558 3.488 352.001 7.674 .000 7937.000 .543 103.162 4.452 518.982 9.499 ‐.202 12183.000 .407 75 846.432 754.743 85.318 0.000 1.170 0.436 30.465 25.300 23.979 Numbers of Firm‐Year Mean Median Std. Deviation Skewness Minimum Maximum Percentiles Panel B ‐ 1988‐2010 Descriptive Statistics Numbers of Firm‐Year Mean Median Std. Deviation Skewness Minimum Maximum Percentiles 38 Source: http://www.doksinet Table 4: Descriptive Statistics of the Final Sample and the Ten‐Industry Sub‐Sample from 1988‐2010 This table presents the descriptive statistics for nine main variables used in the computation of conservatism scores, earnings persistence scores and core‐RNOA for the extended period

1987‐2010. MKT denotes market capitalism of the firm, AT denotes total asset, INVT denotes inventory, LIFR denotes LIFO reserve, EPSPI denotes basic earnings per share, ACC denotes accruals, DP denotes depreciation, XAD denotes advertising expense, and XRD denotes R&D expense. Full Sample 2‐Digits SIC Code Sample Size MKT Mean % of full sample mean Median Standard Deviation AT Mean % of full sample mean Median Standard Deviation 43434 Oil & Gas Food Chemicals Machinery Electronics Transport Instru. Comm. Trade Business 13 2139 20 1916 28 5752 35 4672 36 5910 37 1531 38 4840 48 2081 50‐59 8392 73 6201 2,936 1,764 4,230 4,832 2,554 2,494 4,088 1,039 10,740 1,676 2,171 100% 148 14,141 60% 137 6,475 144% 280 13,670 165% 216 18,860 87% 148 13,653 85% 114 12,080 139% 219 15,151 35% 85 3,733 366% 852 27,763 57% 139 8,772 74% 110 16,064 2,342 1,621 2,817 2,465 1,998 1,726 6,033 783 12,714 1,504 1,138 100% 141.62 11,485 69%

173.55 4,954 120% 369.65 5,862 105% 114.09 9,343 85% 149.71 8,296 74% 108.19 7,274 258% 323.16 26,292 33% 59.79 3,047 543% 1125.48 34,724 64% 220.02 6,078 49% 81.87 6,698 *All levels of accounts are in dollar values are in millions of USD. 39 Source: http://www.doksinet Table 4: Descriptive Statistics of the Final Sample and the Ten‐Industry Sub‐Sample from 1988‐2010 (cont.) Full Sample Oil & Gas 13 2139 Food 20 1916 Chemicals 28 5752 Machinery 35 4672 Electronics 36 5910 Transport 37 1531 Instru. 38 4840 Comm. 48 2081 Trade 50‐59 8392 Business 73 6201 2‐Digits SIC Code Sample Size 43434 INV Mean % of full sample mean Median Standard Deviation 216 100% 12.39 918 47 22% 0.00 210 332 154% 54.50 733 235 109% 9.09 735 246 114% 25.66 1,005 199 92% 17.92 802 636 295% 60.57 2,112 94 43% 10.69 354 154 71% 2.59 467 351 163% 49.04 1,285 41 19% 0.00 461 LIFR Mean % of full sample mean Median Standard Deviation 7.05 100% 0.00 59.73 1.13 16%

0.00 11.54 9.25 131% 0.00 33.64 11.20 159% 0.00 61.83 13.95 198% 0.00 134.71 2.74 39% 0.00 19.46 12.33 175% 0.00 55.09 3.52 50% 0.00 40.43 0.07 1% 0.00 1.02 12.38 176% 0.00 58.64 0.02 0% 0.00 0.61 EPS1 Mean % of full sample mean Median Standard Deviation 0.43 100% 0.33 2.02 0.43 100% 0.21 2.20 1.06 245% 0.80 1.95 0.33 77% 0.09 1.89 0.48 110% 0.38 2.23 0.30 70% 0.26 1.65 1.04 241% 0.87 2.43 0.45 103% 0.29 1.46 0.39 91% 0.32 3.58 0.59 136% 0.59 1.93 0.05 13% 0.11 1.84 ACC Mean % of full sample mean Median Standard Deviation ‐116 100% ‐2.91 863 ‐108 93% ‐10.96 375 ‐97 84% ‐7.42 296 ‐87 75% ‐1.77 477 ‐74 64% ‐2.42 432 ‐94 81% ‐2.40 569 ‐230 197% ‐5.05 1,697 ‐25 21% ‐0.66 148 ‐970 834% ‐38.02 3,079 ‐47 41% ‐3.48 319 ‐55 47% ‐2.69 406 *All levels of accounts are in millions of USD. 40 Source: http://www.doksinet Table 4: Descriptive Statistics of the Final Sample and the Ten‐Industry Sub‐Sample from

1988‐2010 (cont.) Full Sample Oil & Gas 13 2139 Food 20 1916 Chemicals 28 5752 Machinery 35 4672 Electronics 36 5910 T. Equip 37 1531 Instru. 38 4840 Comm. 48 2081 Trade 50‐59 8392 B. Services 73 6201 2‐Digits SIC Code Sample Size 43434 DP Mean % of full sample mean Median Standard Deviation 120 100% 5.65 771 115 96% 13.62 362 105 87% 17.04 211 97 81% 3.56 369 81 68% 5.64 389 92 76% 4.48 396 254 212% 11.81 1,284 30 25% 2.10 128 944 788% 42.27 2,932 56 46% 8.14 217 51 43% 3.60 318 XAD Mean % of full sample mean Median Standard Deviation 93 100% 3.49 352 2 2% 0.12 9 228 246% 18.34 425 181 195% 4.54 593 44 48% 2.56 151 69 74% 0.80 339 274 296% 3.31 822 22 23% 0.86 104 245 265% 12.65 654 67 72% 10.20 178 31 34% 1.24 158 XRD Mean % of full sample mean Median Standard Deviation 103 100% 4.45 519 68 66% 2.64 162 34 33% 8.35 51 196 190% 13.32 764 113 109% 9.00 455 141 136% 7.86 611 282 273% 10.86 969 40 39% 4.36 158 217 210% 9.17 626 1

1% 0.00 30 92 90% 7.53 521 *All levels of accounts are in millions of USD. 41 Source: http://www.doksinet Table 5: Industry Based Regression Results for Conditional Conservatism Models This table presents the results of the AT and AACF multiple regressions performed for each individual industry. β1 is the slope coefficient of the interaction term between dummy variable and proxy for news (return in the case of AT and CFO for the case of AACF). It represents the incremental timeliness of recognition of bad news as compared to good news, and hence the conservatism score. For details of computation and notations of regression variables, please refer to Table 1. AT Regression Equation: ∗ Panel A: AT Regression Results (1988‐1997) Industry 13 20 28 35 36 37 38 48 50‐59 73 β1 0.250 0.700 0.455 0.462 0.505 0.397 0.295 0.227 0.288 0.426 Std. Error 0.237 0.213 0.082 0.090 0.102 0.080 0.039 0.044 0.034 0.058 t 1.053 3.286 5.554 5.138 4.941 4.972 7.633 5.159 8.389 7.304 Panel

B: AT Regression Results (1988‐2010) Sig. Industry 0.293 0.001 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 13 20 28 35 36 37 38 48 50‐59 73 β1 0.384 0.351 0.354 0.361 0.319 0.594 0.281 0.109 0.270 0.441 Std. Error 0.139 0.100 0.043 0.051 0.057 0.089 0.040 0.142 0.029 0.045 AACF Regression Equation: β1 13 20 28 35 36 37 38 48 50‐59 73 0.840 0.138 ‐0.164 0.994 1.208 ‐0.910 0.109 0.062 ‐0.612 1.968 Std. Error 0.061 0.106 0.102 0.090 0.048 0.123 0.077 0.068 0.020 0.415 t 13.866 1.306 ‐1.608 11.015 25.129 ‐7.389 1.418 0.915 ‐31.046 4.743 2.759 3.497 8.267 7.104 5.616 6.684 7.091 0.762 9.342 9.836 Sig. 0.006 0.000 0.000 0.000 0.000 0.000 0.000 0.446 0.000 0.000 ∗ Panel C: AACF Regression Results (1988‐1997) Industry t Panel D: AACF Regression Results (1988‐2010) Sig. 0.000 0.192 0.108 0.000 0.000 0.000 0.156 0.360 0.000 0.000 42 Industry β1 13 20 28 35 36 37 38 48 50‐59 73 0.682 ‐0.985 ‐0.105 0.528 1.055 ‐0.304 0.513 0.941

‐0.390 0.853 Std. Error 0.079 0.061 0.046 0.051 0.019 0.34 0.022 0.024 0.013 0.029 t 8.639 ‐16.039 ‐2.264 10.281 56.911 ‐0.896 23.659 39.808 ‐30.118 29.262 Sig. 0.000 0.000 0.024 0.000 0.000 0.371 0.000 0.000 0.000 0.000 Source: http://www.doksinet Table 6: Summary of Conservatism Scores for Industries This table summarizes all five raw conservatism scores for each industry. Conditional conservatism scores are slope coefficient β1 of regression performed for each industry. Unconditional conservatism scores are the means of all firm‐year scores Industries are then ranked based on the raw scores. Panel A: Raw Conservatism Score and Industry Ranking, 1988‐1997 (1 – least conservative; 10 – most conservative) Industry AT Score Ranking AACF Score Ranking HR Score Ranking NA Score Ranking MTB Score Ranking 13 20 28 35 36 37 38 48 50‐59 73 0.250 0.700 0.455 0.462 0.505 0.397 0.295 0.288 0.288 0.426 1 10 7 8 9 5 4 2 3 6 0.840 0.138 ‐0.164 0.994 1.208

‐0.910 0.109 0.062 ‐0.612 1.968 7 6 3 8 9 1 5 4 2 10 0.01 3.50 4.57 3.58 1.16 1.09 0.89 1.83 0.42 2.67 1 8 10 9 5 4 3 6 2 7 5.35 30.86 10.73 10.81 12.69 19.56 6.84 84.02 11.40 15.77 1 9 3 4 6 8 2 10 5 7 2.51 3.11 4.85 3.40 2.85 2.25 3.96 5.15 2.43 4.48 3 5 9 6 4 1 7 10 2 8 Average Ranking 2.6 7.6 6.4 7.0 6.6 3.8 4.2 6.4 2.8 7.6 Std Dev Ranking 2.6 2.1 3.3 2.0 2.3 2.9 1.9 3.6 1.3 1.5 Panel B: Raw Conservatism Score and Industry Ranking, 1988‐2010 (1 – least conservative; 10 – most conservative) Industry AT‐Score Ranking AACF‐Score Ranking HR‐Score Ranking NA‐Score Ranking 13 20 28 35 36 37 38 48 50‐59 73 0.384 0.351 0.354 0.361 0.319 0.594 0.281 0.109 0.270 0.441 8 5 6 7 4 10 3 1 2 9 0.682 ‐0.985 ‐0.105 0.528 1.055 ‐0.304 0.513 0.941 ‐0.390 0.853 7 1 4 6 10 3 5 9 2 8 0.01 2.03 4.65 2.43 1.30 1.00 1.06 1.14 0.37 2.88 1 7 10 8 6 3 4 5 2 9 28.27 27.56 33.44 42.45 31.53 77.91 13.27 329.15 31.19 34.84 3 2 6 8 5 9 1 10 4 7 43 MTB Score

Ranking 2.45 3.34 4.88 3.09 2.93 2.39 3.65 4.22 2.44 4.18 3 6 10 5 4 1 7 9 2 8 Average Ranking 4.4 4.2 7.2 6.8 5.8 5.2 4.0 6.8 2.4 8.2 Std Dev Ranking 3.0 2.6 2.7 1.3 2.5 4.0 2.2 3.8 0.9 0.8 Source: http://www.doksinet Table 7: Classification Overlap for the Five Conservatism Models at Industry Level This table summarizes the classification overlap among all five unconditional conservatism models. Top three industries with highest score under each conservatism measure are classified as high conservatism, while the lowest scoring three industries are classified as low conservatism. The numbers in the first row of each panel indicate the number of conservatism models that classify a particular industry under the category stated in the panel description. Listing under 0 indicates that the industry is not classified under that category by any of the five conservatism measures. Panel A: High Conservatism 1988‐1997 Number of Conservatism Models Agree in Classification 5 4 3 20 35

2 28 36 48 73 1 37 0 13 38 50 4 13 50 3 2 37 28 38 1 48 0 20 35 36 73 3 48 2 28 37 35 1 13 36 0 10 38 50 3 13 37 2 20 38 1 48 0 28 35 36 73 Industry Panel B: Low Conservatism 1988‐1997 Number of Conservatism Models Agree in Classification Industry 5 Panel C: High Conservatism 1988‐2010 Number of Conservatism Models Agree in Classification Industry 5 4 73 Panel D: Low Conservatism 1988‐2010 Number of Conservatism Models Agree in Classification Industry 5 44 4 50 Source: http://www.doksinet Table 8: Classification Overlap among Unconditional Conservatism Measures within Each Industry This table summarizes the classification overlap at firm‐year level for individual industries for both 1988‐1997 and 1988‐2010. Top 30% firm‐years of each industry ranked under each conservatism measure are classified as high conservatism, while the lowest 30% firm‐years of each industry ranked are classified as low conservatism. The firm‐year overlap is the

number of firm‐years that is classified under the same category by all three unconditional conservatism measures % Overlap is computed as firm‐year overlap/30% of total firm‐years classified under the industry. It is important to note that different industry portfolios have different sample sizes, and hence % overlap rather than firm‐year overlap should be used for comparison between industries. Industry Total Firm‐ years 1988‐1997 High Conservatism 13 20 28 35 36 37 38 48 50 1045 901 2530 2251 2682 707 2278 910 4091 Firm‐year Overlap 91 13 65 56 80 15 80 11 69 73 2536 62 Total Firm‐ years Low Conservatism % Overlap 29% 5% 9% 8% 10% 7% 12% 4% 6% Firm‐year Overlap 77 14 110 74 122 27 88 42 109 % Overlap 25% 5% 14% 11% 15% 13% 13% 15% 9% 8% 91 12% 45 1988‐2010 High Conservatism Low Conservatism 2139 1916 5752 4672 5910 1531 4840 2081 8392 Firm‐year Overlap 153 34 178 131 180 45 142 23 163 % Overlap 24% 6% 10% 9% 10% 10% 10% 4% 6% Firm‐year

Overlap 178 71 251 131 226 71 169 52 231 % Overlap 28% 12% 15% 9% 13% 15% 12% 8% 9% 6201 152 8% 180 10% Source: http://www.doksinet Table 9: HR Q‐Score Comparison This table summarizes the statistics of filtered Q‐Score (Q HR Score), and the Q‐Score computed and published in Penman and Zhang’s paper (2002). This study replicates that of Penman and Zhang (2002). However, data period for this part of analysis differs from that of Penman and Zhang (2002). Filtered Q HR Score Number of Firm‐years Q Score (Table 1) 33,943 29,796 1988‐2010 1975‐1997 0.350 0.099 95 1.858 0.219 75 0.139 0.059 50 ‐0.011 0.009 40 ‐0.032 0.000 25 ‐0.101 ‐0.010 10 ‐0.205 ‐0.046 5 ‐0.282 ‐0.075 Data Period Mean Percentiles Penman & Zhang 46 Source: http://www.doksinet Table 10: Summary of Results of Annual Cross‐Sectional Regressions This table summarizes the regression results replicating Penman and Zhang (2002) annual

cross‐sectional regressions for the period 1976‐1996. The t‐statistic is calculated based on a one‐sample t‐test of annual coefficients obtained against the hypothesis that the mean of coefficients equals to zero. For this study, 22 annual cross‐ sectional regressions were performed for each of the Q‐Scores for the period 1988‐ 2009. Following Penman and Zhang, three similar one‐sample t‐tests were performed for each of the Q‐scores testing the mean of 22 coefficients for each case. * Significant at 1%, and * significant at 5%. Earnings Persistence Equation: RNOAt+1 = α0 + α1RNOAt + α2Qt + et+1 Penman and Zhang Q HR (Table 3, Panel A) Intercept RNOA coefficient Q coefficient Q HR Intercept RNOA coefficient Q coefficient Q MTB Intercept RNOA coefficient Q coefficient Q NA Intercept RNOA coefficient Q coefficient Mean 0.016* 0.800* 0.096* First Quartile 0.009 0.782 0.054 Median 0.018 0.813 0.103 Third Quartile 0.022 0.846 0.136 Mean 0.029* 0.284* 0.027*

First Quartile 0.008 0.188 0.002 Median 0.044 0.256 0.030 Third Quartile 0.060 0.394 0.050 Mean 0.036* 0.263* 0.018 First Quartile 0.022 0.139 0.007 Median 0.037 0.255 0.018 Third Quartile 0.073 0.343 0.035 First Quartile 0.019 0.141 <0.000 Median 0.040 0.255 <0.000 Third Quartile 0.081 0.369 <0.000 Mean 0.028 0.263 <0.000 47 Source: http://www.doksinet Table 11: Summary of Results of Pooled Regressions This table summarizes the results of the regression for the filtered sample as a whole. Individual regressions were run for each of the Q‐Scores using the Penman and Zhang’s earnings persistence regression equation: RNOAt+1 = α0 + α1RNOAt + α2Qt + et+1. * Significant at 1%, and significant at 5%. Panel A: Overall Regression Result for Q HR Variable Coefficient t‐Statistic C 0.0208* 2.6373 CRNOA T 0.2491* 33.7165 Q HR 0.0185* 6.4362 F‐statistic 574.7051* Panel B: Overall Regression Result for Q MTB Variable Coefficient t‐Statistic

C 0.0240* 3.0414 CRNOA T 0.2368* 33.4219 Q MTB 0.0118* 3.3847 F‐statistic 559.2320* Panel C: Overall Regression Result for Q NA Variable Coefficient t‐Statistic C 0.0256* 3.2547 CRNOA T 0.2350* 33.2323 Q NA 0.0002* 2.2110 F‐statistic 555.8412* 48 Source: http://www.doksinet Table 12: Summary of Results of Industry Stacked Regression The industry coefficients were obtained by running a stacked regression using the Penman and Zhang’s earnings persistence regression equation: RNOAt+1 = α0 + α1RNOAt + α2Qt + et+1. Pooling was performed by assigning dummy variables to nine out of ten selected industries. Business (73) was the base industry and its regression coefficients were subsumed in the constant term A separate stacked regression was performed for each of the three Q‐Scores. * Significant at 1%, and significant at 5%. Panel A: Industry Regression Result for Q HR Variable Coefficient t‐Statistic Panel B: Industry Regression Result for Q MTB

Variable Coefficient t‐Statistic Panel C: Industry Regression Result for Q NA Variable Coefficient t‐Statistic Constant 0.0219* 2.7824 Constant 0.0235* 2.9748 Constant 0.0257 CRNOA T 0.2501* 33.4517 CRNOA T 0.2335* 32.862 CRNOA T 0.2350* Q HR*D 13 ‐2.5372* ‐3.0362 Q MTB*D 13 ‐0.0688 Q NA*D 13 0.0002 0.2867 Q HR*D 20 0.0398* 2.2216 Q MTB*D 20 5.6679 Q NA*D 20 0.0002 0.7643 Q HR*D 28 0.0131* 2.7056 Q MTB*D 28 ‐0.0132 ‐1.8114 Q NA*D 28 0.0002 0.8958 Q HR*D 35 0.0288* 2.9398 Q MTB*D 35 ‐0.0174 ‐1.4938 Q NA*D 35 0.0002 0.8218 Q HR*D 36 0.0677* 8.5352 Q MTB*D 36 0.0182 1.8271 Q NA*D 36 0.0001 0.6199 Q HR*D 37 0.0033 0.1307 Q MTB*D 37 0.0070 0.2621 Q NA*D 37 0.0002 0.5981 Q HR*D 38 0.0237 1.7581 Q MTB*D 38 0.0140 1.4343 Q NA*D 38 0.0001 0.4153 Q HR*D 48 0.0002 0.0113 Q MTB*D 48 0.0027 0.2065 Q NA*D 48 0.0002 0.8353 Q HR*D 50 ‐0.0393* ‐2.4151 Q MTB*D 50 0.0364* 3.2794 Q NA*D 50

0.0003 1.2764 F‐statistic 121.8645* F‐statistic ‐0.0039 0.1027* 116.1929* 49 F‐statistic 111.1905* 3.2654 33.2289 Source: http://www.doksinet Figure 1 : Digrammatic Representation of the Factors which Contribute to the Conservatism of a Company This figure summarizes the major contributions to the conservatism of a firm. ↓Revenue Delay recognition of revenue:  Revenue recognition methods which have time dimension: cost recovery, percentage of completion method  More stringent recognition and measurement criteria – ↑Expenses = ↓Earnings Accelerated recognition of expenses:  Write-offs  Inventory (lower-of-cost or market value)  Accounts Receivable  Tangible assets o Depreciation methods used o Impairment charges  Intangible assets o Amortization methods used o Impairment charges  Provisions (creation of liability)  Provision for warranty  Accrued expenses (leaves, staff benefits etc)  Contingent liabilities 

Capitalization versus expense  Future benefits generating expenses (R&D, advertising, brand names, patent expenses, interest on construction Net Effect is that annual earnings and cash flows become out of synchronization. 50 Source: http://www.doksinet Figure 2: Conservatism Score Ranking for Industries This figure shows the conservatism rank computed for each industry for each of the five conservatism measurement models for the period 1988‐1997 and 1988 ‐ 2010. The y‐ axis indicates the rank whereas the x‐axis indicates the measurement model used to compute the score. Each series shows the ranks of one industry under each conservatism model. The legend indicates the code of the industry represented by each series. Industries are ranked from 1‐10 with 1 being the least conservative and 10 being the most conservative. Panel A : Conservatism Score Ranking for Industries for the Period 1988‐1997 10 13 9 20 8 28 7 6 35 5 36 4 37 3 38 2 48 1 50‐59

AT Score AACF Score HR Score NA Score MTB Score 73 Panel B : Conservatism Score Ranking for Industries for the Period 1988‐2010 10 13 9 20 8 28 7 6 35 5 36 4 37 3 38 2 48 1 50‐59 AT Score AACF Score HR Score NA Score 51 MTB Score 73 Source: http://www.doksinet Figure 3: Classification by Conditional Conservatism Models at Industry Level This figure consists of four Venn diagrams of the classification results arranged into two panels. Panel A (B) presents the results obtained for 1988‐1997 (1988‐2010) Top three industries ranked under each conditional conservatism measure are classified as high conservatism, while the lowest three industries are classified as low conservatism. Overlap in classification is expressed as the overlap between the circles in the Venn diagram. Only conditional conservatism models are included in this analysis Panel A: Industry Classification for the years 1988‐1997 High Conservatism Low Conservatism AACF AACF Panel

B: Industry Classification for the years 1988‐2010 High Conservatism Low Conservatism AACF AACF 52 Source: http://www.doksinet Figure 4: Classification by Unconditional Conservatism Models at Industry Level This figure consists of four Venn diagrams of the classification results arranged into two panels. Panel A (B) presents the results obtained for 1988‐1997 (1988‐2010) Top three industries ranked under each unconditional conservatism measure are classified as high conservatism, while the lowest three industries are classified as low conservatism. Overlap in classification is expressed as the overlap between the circles in the Venn diagram. Only unconditional conservatism models are included in this analysis. Panel A: Industry Classification for the years 1988‐1997 High Conservatism Low Conservatism Panel B: Industry Classification for the years 1988‐2010 High Conservatism Low Conservatism 53 Source: http://www.doksinet Figure 5 : Firm‐year Classification

Using Unconditional Conservatism Models This figure consists of four Venn diagrams. Diagrams Panels A and B (C and D) present the results obtained for the period 1988‐1997 (1988‐2010). Top 30% firm‐years (13,030 firm‐ years) ranked under each conservatism measure are classified as high conservatism, while the lowest 30% firm‐years (13,030 firm‐years) are classified as low conservatism. Percentage in parenthesis indicates the percentage of overlap (number of overlap firm‐ years / 30% of total firm years). Only unconditional conservatism models are included in this analysis. Period of 1988 to 1997 Panel A : High Conservatism Panel B : Low Conservatism Period of 1988 to 2010 Panel D : Low Conservatism Panel C : High Conservatism 54