Conditional Conservatism in U.S. High- and Low- Technology Firms 1. Khalifa Mariem Ph.D candidate Manouba University

Transcription

1 Conditional Conservatism in U.S. High- and Low- Technology Firms 1 Khalifa Mariem Ph.D candidate Manouba University Samir Trabelsi 2 Associate Professor of Accounting Brock University Hamadi Matoussi Professor of Finance Manouba University Abstract: In this study, we investigate whether U.S. high-technology firms are more or less conditionally conservative relative to low-technology firms. If U.S. high-tech firms are required to expense immediately all R&D costs according to the accounting standard SFAS 2, which reflects unconditional conservatism, then in case of bad news (conditional conservatism), these firms cannot write-off their R&D costs. We expect that high-tech firms are less conditionally conservative since unconditional conservatism dampens the timeliness of accounting income. Our findings are consistent with the prediction that high-technology firms are less conditionally conservative than low-technology firms. The level of conditional conservatism increases with the level of leverage regardless of the technology level of the firm. Only low-technology firms are more conditionally conservative when they face higher auditor litigation risk. We find that neither in high-technology firms nor in low-technology firms, taxation affects conditional conservatism. Keywords: Conditional conservatism; High-technology; Low-technology; Contracting; Litigation risk; Taxation. JEL Classification: C23, L63, L65, L86, L96, M41, O Introduction Accounting conservatism is defined by Basu (1997, p. 7) as the accountant s tendency to require a higher degree of verification to recognize good news as gains than to recognize bad news as losses. According to Huijgen and Lubberink (2005), Sivakumar and Waymire 1 We thank Mohamed Ayadi; Sudipta Basu; Katherine Schipper; Skander Lazrak and participants at the European Business Research Conference and at Brock University workshop for their valuable comments. 2 Corresponding author. 1

2 (2003) and Givoly et al. (2007), conservatism is the most influential principle of valuation in accounting. Many studies demonstrate that the degree of accounting conservatism is higher in U.S. firms than in other countries (Basu, 1997; Ball et al., 2000; Givoly and Hayn, 2000; Holthausen and Watts, 2001; Sivakumar and Waymire, 2003; Ryan and Zarowin, 2003). Recent empirical studies investigate the level of accounting conservatism within high- and low-technology firms (Kwon et al., 2006; Chandra, 2011). However, those studies find mixed results when they use the regression coefficients from the income timeliness model in Basu (1997). The primary research objective of this paper is to extend the Kwon et al. (2006) and Chandra (2011) studies by comparing the level of conditional conservatism between high and low-technology firms in the U.S. In Watts (2003), there are four main explanations for accounting conservatism: contracting, litigation, taxation and regulation. Recent research by Qiang (2007) and García-Lara et al. (2009) examine the effect of these four determinants on conditional and unconditional forms of conservatism. Therefore, the second research objective of this paper is to extend the analyses of Qiang (2007) and García-Lara et al. (2009) by investigating the drivers of the cross-sectional difference in conditional conservatism in high- and low-technology firms. An investigation into the degree of conditional conservatism in high- and low-technology firms, as well as testing the effect of each determinant on conservative accounting in these two technology sectors, is worthwhile for several reasons. First, anecdotal evidence such as the fall of stock prices of Apple Inc., Microsoft Corp., Intel Corp., and Hewlett-Packard Co. on November 16, 2012 in The Globe and Mail may be explained by their overpricing or underpricing. Second, the diversity of the characteristics of high-technology firms leads us to investigate whether they would adopt more or less accounting conservatism. In fact, hightechnology firms have higher growth opportunities and more volatile stock prices and face higher shareholder litigation risk (Francis et al., 1994; Jones and Weingram, 1996; Johnson et al., 2001). Also, these firms are characterized by high levels of information asymmetry (Aboody and Lev, 2000; Chan et al., 2001, Barth et al., 2001; Ljungqvist and Wilhelm, 2003). Moreover, high-technology firms play an important role in U.S. economy in the last decades. According to Chan et al. (2001), the technology-based firms and the pharmaceutical industry together accounts for 40% of the value of the S&P 500. That is why it is very crucial to examine whether high-technology firms are more or less conditionally conservative in their financial reports relative to low-technology firms. Finally, prior academic research on accounting conservatism in the high- and low-technology sector shows mixed evidence when 2

3 using Basu's model. Kwon et al. (2006) demonstrate that the level of conditional conservatism is higher in high-technology firms than in low-technology ones. However, Chandra (2011) finds that the average yearly value of Basu's model intercept is significantly lower for the technology sample compared to non-technology. He also finds that the time-series average incremental slope coefficient for bad news in Basu's (1997) earning-return regression model shows lower average conditional conservatism for technology firms than non-technology ones. Thus, Chandra (2011) concludes that Basu's model provides evidence of greater unconditional conservatism for technology firms, but it is unable to document differences in conditional conservatism. Therefore, it is worth focusing on the conditional form of conservatism in these two technology sectors. Researchers proposed four main explanations for the market demand of accounting conservatism: contracting, litigation, regulation, and taxation (Watts, 2003; Qiang, 2007 and García-Lara et al., 2009). Then, investigating the market demand factors, which may explain the cross-sectional differences in conditional conservatism between high- and low-technology firms, will help to further our understanding of which of these economic determinants will generate motivations to adopt conditional conservatism in these two sectors. Using the model of Basu (1997) to a large sample of U.S. firms operating in high- and lowtechnology sectors as in the Francis and Schipper (1999) for the period of 1974 to 2010, we show that high-technology firms are less conditionally conservative than low-technology firms. This lower conservatism level in the high-technology firms is manifested in significantly lower asymmetric timeliness of bad vs. good news for the high-technology sample which is relative to the low-technology sample. Further, we provide evidence that conditional conservatism and leverage are positively associated in both high and lowtechnology firms. Moreover, we find that conditional conservatism increases with the level of auditor litigation risk, which only appears in low-technology firms. Finally, the findings show that taxation does not induce conditional conservatism in both technology samples. This study contributes to prior literature in several ways. First of all, this paper extends the empirical research by Kwon et al. (2006) and Chandra (2011) by showing that conditional conservatism is measured by asymmetric timeliness of bad vs. good news, which is lower in high-technology firms than low-technology ones. Prior studies compare the degree of accounting conservatism in high and low-technology firms and find contradictory results. By demonstrating that accounting conservatism is less pronounced in high-technology firms, this study confirms that these firms are less conditionally conservative than low-technology ones. 3

4 Second, our study uses the C-score measure developed by Khan and Watts (2009) to investigate the level of conditional conservatism in high and low-technology firms. Third, despite the large amount of literature on accounting conservatism and its determinants, to the best of our knowledge, there are not any empirical studies treated with the determinants of the demand for accounting conservatism in accordance to the technological level of the firms. Thus, the originality of our research is to identify the market demand factors which may explain the cross-sectional difference in conditional conservatism between high- and lowtechnology firms. Third, the results of this paper will be particularly useful for the financial analysts. They will encourage them to concentrate more on the analysis of firms operating in the high- and low-technology sector with the aim not to mislead investors. These last years, in fact, have seen the nature of the activities of firms evolve a lot, in particular since the development of high technologies, investors are brought in more and more frequently to assess the high-technology firms value of growth opportunities, which are characterized by a strong information asymmetry. So, these analysts have to take into account the technology levels before proceeding in comparison to the degree of conservatism between firms. Finally, the importance of the role of accounting conservatism is seen in reducing auditor litigation risk in low-technology firms; the results of this study will have implications on the standard setters who intend to abandon accounting conservatism in favor of the neutrality. In fact, this work can lead to the Financial Accounting Standards Board (FASB) and the International Accounting Standards Board (IASB) to strengthen their position or to revise their position in regards to the exclusion of accounting conservatism from the qualitative characteristics of accounting information. The remainder of this paper is organized as follows: Section 2 presents a review of relevant literature. The third section develops the hypotheses. The fourth section describes the research design. The fifth section presents the sample selection and the empirical results. The sixth and final section provides a summary and conclusions of our findings. The appendix presents variable definitions. 2. Literature Review The extant literature provides strong evidence of the various strongly conservatism degrees. Ball et al. (2000) find that higher conservatism among firms in common-law countries. As seen in Ahmed et al. (2002), greater conservatism is among firms where conflicts between bondholders and shareholders over dividend policy are higher. Sivakumar and Waymire's 4

5 (2003) evidence indicates that firms under more intense rate regulation exhibit greater conditional conservatism. Ball and Shivakumar (2005) examine accounting conservatism in public vs. private U.K. firms. They document that conservatism is more pronounced among public firms. Recently, LaFond and Watts (2008) show that firms facing more information asymmetry between insider and outsider equity investors are more conservative. Chi and Wang (2010) extend LaFond and Watts (2008) findings, which further confirm that accounting conservatism is positively related to the level of information asymmetry. Moreover, these authors suggest that the standard setters, either the FASB or IASB, should not exclude conservatism from the qualitative characteristics of accounting information since it is an efficient tool to address the moral hazard problem arising from information asymmetry. Although there is ample evidence of the different levels of conservatism across countries and between firms, very few studies investigate the relation between the high- and low-technology sector and the level of the accounting conservatism. In fact, the papers of Chandra (2011) and Kwon et al. (2006) constitute, in our knowledge, the only studies to have examined the degree of accounting conservatism within the high and low-technology firms in a detailed way. In their paper, Kwon et al. (2006) examine the level of accounting conservatism on a sample of 2,728 high-technology firms and 984 low-technology firms as shown in Francis and Schipper (1999) during the period 1990 to The findings of Kwon et al. are based on comparisons of cumulative non-operating accruals, while the regression coefficients from the income timeliness models in Basu (1997), the distribution of earnings, and discretionary accruals between the two groups are consistent with a higher level of accounting conservatism in high-technology firms as compared to low technology firms. More particularly, these results indicate that bad news is registered in the earnings of financial statements more rapidly than good news in high-technology firms versus low-technology firms. Kwon et al. (2006, p ) identify five reasons to explain the positive relation between the high-technology sector and accounting conservatism. Among these explanations, they, first, indicate that accounting standards require high-technology firms to apply more conservative standards, such as SFAS 121 on asset impairments, SFAS 5 on contingencies, SFAS 2 which requires immediate expensing of most R&D costs, SFAS 86 on software development costs and AICPA SOP 97-2 which requires deferral of certain software revenue. Second, a higher number of growth opportunities and an increased risk due to volatile stock prices made hightechnology firms more vulnerable to shareholder litigation. Besides, high-technology firms 5

6 attracted more attention from financial analysts because of their high growth opportunities. So, they are likely more incited to report conservative accounting numbers in their financial statements. Moreover, high-technology firms are motivated to be more conservative in order to signal that they are able to meet investors expectations about future growth. Finally, during the 1990s, which are characterized by the irrational exuberance in the U.S. stock market and the formation of record budget surpluses from government, firms operating in the high-technology sector were motivated to become more conservative in order to smooth earnings over time. Furthermore, the research conducted by Chandra (2011) for the period between 1982 to 2001 shows also that technology firms are characterized by higher levels of income conservatism. Their further analysis suggests that technology firms higher conservatism results primarily from conservative accounting rules for R&D expensing rather than shareholder litigation risk. However, results based on the asymmetric timeliness coefficient from Basu s model revealed that high-technology firms have a lower AT coefficient in comparison to non-technology firms. He finds also that the Basu's model intercept is consistent with higher unconditional conservatism for technology firms. Thus, Chandra (2011) concludes that news-independent conservatism related to R&D expensing is the primary determinant of income conservatism for U.S. technology firms. He explains this lower incremental slope coefficient for technology firms by two reasons. First, he argues that since stock returns contain information on changes in growth opportunities, which is not captured by earnings irrespective of level of conservatism, the Basu incremental slope is lower for firms with higher variability in growth expectations. Second, he considers that the Basu slope coefficient is positively associated with debt, but the high-technology sample has lower debt. That is why the Basu slope coefficient is lower for technology firms. Thus, we notice that there have been debates on whether the asymmetric timeliness coefficient from Basu s regression is lower or higher for high-technology firms. So why, do we revisit the question of whether conditional conservatism varies with the technological level of firms in this study? We will test if conditional conservatism based on the Basu s model is greater for high or low-technology firms. Like previous analyses (Kwon at al., 2006; Chandra, 2011), we also use the classification of Francis and Schipper (1999). 6

7 3. Hypotheses Development Conditional conservatism in high-tech firms Few studies examined whether the degree of accounting conservatism varies across technology and non-technology firms and provide mixed results. Kwon et al. (2006) use a sample of 2,728 high-technology firms and 984 firms of low technology during the period of 1990 to 1998 and document that high-technology firms are more conservative than lowtechnology ones. Similarly, Chandra (2011) shows that technology firms are characterized by higher levels of income conservatism. In contrast, Chandra (2011) finds lower asymmetric timeliness in high-technology firms in comparison to non-technology ones. He concludes that news independent conservatism associated to R&D expensing is the primary determinant of income conservatism for U.S. technology firms. These findings indicate that accounting conservatism level differs between high and low-technology firms. But, it is unclear if the asymmetric timeliness of economic losses vs. economic gains is more pronounced in high- or low-technology firms. Some accounting standards reflect that asymmetric timeliness of bad vs. good news should be less pronounced in high-technology firms relative to low-technology firms. For instance, American firms are required to expense immediately all research and development costs under SFAS 2 and not to activate them, which lead to greater levels of unconditional conservatism and thus, gives place to less future conditional conservatism depreciation due to the pre-emptive role of ex-ante conservatism on conditional conservatism. High-technology firms are more likely to be affected by this rule than low-technology firms because they have more R&D investments. Thus, these high-technology firms are more likely to have lower earnings levels relative to low-technology firms (Lev et al., 2005; Healy et al., 2002). Beaver and Ryan (2005) explain that in case of bad news (conditional conservatism), the firm cannot write-off its R&D costs when it does not capitalize on them (unconditional conservatism). This means that unconditional conservatism is negatively associated with the conditional form of conservatism and the ex-ante conservatism pre-empts the ex-post form. In fact, García-Lara et al. (2009) find that firms with greater levels of unconditional conservatism are less conditionally conservative in their reports. They consider that unconditional conservatism dampens the timeliness of accounting incomes. Moreover, Giner and Rees (2001), Basu (2001), Pope and Waller (2003), and Beaver and Ryan (2005) show that unconditional conservatism pre-empts the conditional form. García-Lara and Mora (2004), Pae et al. (2005), Gassen et al. (2006), Givoly et al. (2007), and Roychowdhury and Watts (2007) also provide 7

8 evidence of a negative relationship between conditional and unconditional conservatism. Consequently, U.S. high-technology firms will exhibit more unconditional conservatism in their financial reports since they are required to apply some conservative accounting standards, such as SFAS 2 on (R&D) costs. By not activating the research and development costs it will be impossible for high-technology firms to ex-post depreciate these intangible assets (conditional conservatism). Given the pre-emptive action of unconditional conservatism on conditional conservatism (Giner and Rees, 2001; Basu, 2001; Pope and Waller, 2003; Beaver and Ryan, 2005), this higher level of unconditional conservatism under SFAS 2 in high-technology firms leads to a less conditional degree of conditional conservatism as compared to low-technology firms. In other words, the membership of firms in the high-technology sector is more likely to increase the unconditional conservatism and to conversely the asymmetric timeliness of bad vs. good news. The above reasoning provides us with our first hypothesis: HYPOTHESIS 1. Conditional conservatism in high-tech firms is lower compared to low-tech firms. The determinants of the lower level of conditional conservatism in high-tech firms In Hypothesis 1, we expect that conditional conservatism is less pronounced in hightechnology firms than in low-technology firms. The next question is what drives this crosssectional difference. In order to avoid agency costs (Jensen and Meckling, 1976), debt holders and the firm contract are based on accounting numbers. The managers have incentives to introduce bias to accounting measures, especially when their compensation is based on accounting numbers. Under contracting perspective, conservative accounting is a means of addressing moral hazard caused by the parties to the firm having asymmetric information, asymmetric payoffs, limited horizons and limited liability (Watts, 2003a, p. 209). Also, debt contracting is considered as the most important explanation for conditional conservatism (Holthausen and Watts, 2001; Watts, 2003; Guay and Verrecchia, 2006, and Ball, Robin and Sadka, 2008) and it improves contracting efficiency (Ball and Shivakumar, 2005). Lenders have incentives to protect themselves, which is why they require that the managers must recognize losses in a timely fashion (Beatty et al., 2008). According to Watts (2003a), LaFond and Watts (2008) and Khan and Watts (2009), high levered firms are more likely to demand more conditional conservatism levels. Highly levered firms have more agency conflicts between lenders and shareholders. Firms can minimize these expected agency costs by recognizing bad news as early as possible. In addition, many studies have demonstrated that 8

9 contracting induces only the conditional form of conservatism (Qiang, 2007; García-Lara et al., 2009). Unconditional conservatism does not employ new information (Basu, 2005), and introduces noise to payoffs to contracting parties (Ball and Shivakumar, 2005). Overall, we hypothesize that leverage is expected to induce only conditional conservatism. Khan and Watts (2009, p. 135) claim that financially distressed firms are more likely to be sued, and the likelihood of financial distress is increasing in leverage, suggesting a higher litigation demand for conservatism from more levered firms. High-technology firms have more growth opportunities relative to assets-in-place. According to Smith and Watts (1992), there is a positive association between growth options and agency costs. These firms with more growth options are likely to have fewer debts. In other words, given that hightechnology firms have more growth opportunities and fewer tangible assets, then they have lower levels of leverage and thus, less agency conflicts of debts. Since conditional conservatism constitutes an efficient corporate governance tool to reduce agency costs (Watts, 2003a), then debt holders are less likely to demand conditionally conservative from hightechnology firms. However, because low-technology firms have greater levels of leverage, debt holders demand from them to incorporate bad news as losses in earnings in a timely fashion than good news such as gains. Then, low-technology firms respond to their demand by being more conditionally conservative in their financial statements. Therefore, we propose the following hypothesis: HYPOTHESIS 2. The lower level of conditional conservatism in high-tech firms is driven by their lower level of leverage. Another possible explanation of conditional conservatism is the litigation risk. Lawsuits against shareholders and auditors are more likely to happen when earnings and net assets are overstated (Kellog, 1984; St. Pierre and Anderson, 1984). If this is the case, then, both managers and auditors are motivated to be conservative in their financial reports in order to reduce litigation costs (Chung and Wynn, 2008). According to Qiang (2007) and García-Lara et al. (2009), litigation generates incentives for both conditional and unconditional conservatism. Furthermore, the degree of conditional conservatism increases with higher probability litigation with an auditor (LaFond and Watts, 2008). To reduce litigation costs, auditors tend to be more conservative and publish unfavorable audit opinions, increase their auditing fees (Pratt and Stice, 1994), or abandon relationships with clients with more risk (Krishnan and Krishnan, 1997). Also, Basu (2001), Basu (1997), Holthausen and Watts (2001), and Khan and Watts (2009) argue that conditional conservatism has increased 9

10 (decreased) with the increase (decrease) of litigation risk. Basu et al. (2001) argue that Big Eight auditors have incentives to be more conservative compared to non-big Eight auditors because of their greater legal liability exposure. Their findings are consistent with the Big Eight auditors report more (less) rapidly bad (good) news comparing to non-big Eight auditors. Moreover, Krishnan (2007) documents that large audit firms issue more conditionally conservative reports in order to decrease their litigation risk exposure. In case of audit failure, larger auditors are more likely to suffer from losses. Smith and Watts (1992) document that high-technology firms are more likely to have a higher probability of litigation. In fact, the higher growth opportunities and increased risk faced by technology firms makes them more vulnerable to litigation (Francis et al., 1994; Jones and Weingram, 1996; Skinner, 1997; Johnson, Kasznik and Nelson, 2001). Hence, since high-technology firms face greater litigation risk, there is a higher litigation demand for conditional conservatism from hightechnology firms. On the other hand, low-technology firms have more tangible assets, so the risk of overvaluation is higher. In the case of overestimation of net assets, the litigation risk is greater in low-technology firms, which suggests that they are more likely to be more conservative. This leads to the following null hypotheses: HYPOTHESIS 3. The lower level of conditional conservatism in high-tech firms is not driven by auditor litigation risk level. The final possible reason that may explain the difference in the level of conditional conservatism is taxation. Shackelford and Shevlin (2001) document that financial accounting choices are affected by tax considerations. According to Ball (2001), Ball and Shivakumar (2005), Basu (2005) and Qiang (2007), taxation induces only the unconditional form of conservatism. However, García-Lara et al. (2009) find evidence consistent with both conditional and unconditional conservatism, which are used by managers as tools to reduce the present value of taxes and, then, to increase firm value. They argue that conditional conservatism is used to shift current income from periods with high expected tax rates to periods with lower expected tax rates. Consequently, taxation generates incentives for conservatism in financial reports. In highly profitable firms, managers prefer to be more conservative to reduce the present value of taxes and, thus, increasing the value of the firm. Smith and Watts (1992) argue that firms with more growth options relative to assets-in-place are more likely to have lower taxable earnings. In other words, it is low-technology firms that are incentivized to report lower book earnings in order to reduce tax payments rather than high-technology firms, suggesting a higher taxation demand for conditional conservatism 10

11 from low-technology firms. Consequently, the demand of high-technology firms for conditional conservatism is lower. Accordingly, we hypothesize the following: HYPOTHESIS 4. The lower level of conditional conservatism in high-tech firms is not driven by their taxable earnings. 4. Research design Conditional conservatism Conditional conservatism can be measured by the Basu coefficient. Ryan (2006) argues that despite some critics and limitations, the Basu s measure of asymmetric timeliness continues to be the best measure of conditional conservatism available and the most popular measure of accounting conservatism. In addition, Ball et al. (2011) confirm that the Basu model is conceptually considered more relevant compared to other measures of conservatism. Basu (1997, p. 7) defines conservatism as the accountant s tendency to require a higher degree of verification in order to recognize good news as gains than to recognize bad news as losses. Basu supposes that earnings capture bad news in a timely fashion than good news. Specifically, Basu (1997) estimates the following model: X it / P it-1 = β 0 + β 1 D it + β 2 R it + β 3 D it * R it + ε it (1) Where X it is earnings per share (Compustat # 58) divided by the beginning of the fiscal year price (Compustat # 199); R it is the annual stock return of the firm, measured compounding twelve monthly Center for Research in Security Prices (CRSP) stock returns ending three months after the fiscal year-end t; D it is a dummy variable that equals to 1 if returns are negative, and 0 otherwise. In this model, the coefficient β 3 measures the level of conservatism and it is expected to be positive and significant. In addition, we use the measure proposed by Khan and Watts (2009) in order to proxy conservatism in our sample. This measure is an extension of the Basu's model and estimates the level of conservatism per firm and per year. According to Khan and Watts (2009), the conservatism score is a function of firm-specific characteristics: firm size, leverage, and the market-to-book ratio. First, we use the following annual cross-sectional Fama MacBeth regression to estimate C_score and G_score: X it / P it-1 = β 0 + β 1 D it + R it (µ 0 + µ 1 Size it + µ 2 MTB it + µ 3 LEV it ) + D it *R it (ʎ 0 + ʎ 1 Size it + ʎ 2 MTB it + ʎ 3 LEV it ) + (δ 1 Size it + δ 2 MTB it + δ 3 LEV it + δ 4 D it *Size it + δ 5 D it *MTB it + δ 6 D it *LEV it ) + ε it (2) 11

12 Where X it is earnings per share (Compustat # 58) deflated by the beginning of the fiscal year price (Compustat # 199); R it is the annual stock return of the firm, measured compounding twelve monthly CRSP stock returns ending three months after the fiscal year-end t; D it is a dummy variable that equals to 1 if returns are negative, and 0 otherwise; Size is the natural log of market value of equity (Compustat # 25 * Compustat # 199); MTB is the market-to-book ratio [(Compustat # 25* Compustat # 199) / Compustat # 60]; LEV is leverage, measured as the sum of long-term and short-term debt (Compustat # 9 + Compustat #34) scaled by the total numbers of assets (Compustat # 6). Then, we calculate C_Score and G_Score for each firm-year as follows: C_Score = ʎ 0 + ʎ 1 Size it + ʎ 2 MTB it + ʎ 3 LEV it (3) G_Score = µ 0 + µ 1 Size it + µ 2 MTB it + µ 3 LEV it (4) Where C_Score reflects the incremental timeliness of bad news; G_Score reflects the timeliness of good news; Size is the natural log of the market value of equity; MTB is the market-to-book ratio; LEV is leverage, measured as the sum of long-term and short-term debt deflated by total assets. We expect that U.S. firms are conservative in general. Then, we expect that the mean of C_Score will be higher than the mean of G_Score. Conditional conservatism in high- and low-technology sector Hypothesis 1 examines the level of conditional conservatism in high- and low-technology firms. We test it by comparing the Basu model between the sample of high-technology and the sample of low-technology firms. The Basu model is as follows: X it / P it-1 = β 0 + β 1 D it + β 2 R it + β 3 D it * R it + ε it Where X it, R it and D it as previously defined. We estimate this model by using a pooled methodology during the sample period of 1974 to The coefficient β 3 reflects the asymmetric timeliness in the recognition of bad vs. good news. Hypotheses 1 predicts the coefficient of D it *R it (β 3 ) will be lower and significant for the high-technology sample, suggesting that the level of conditional conservatism is lower for high-technology firms than low-technology firms. Second, the use of C_Score as a metric of conservatism is to compare the level of conditional conservatism between high- and low-technology firms. We expect that the mean of C_Score is significantly lower for the high-technology sample than for the low-technology 12

13 sample, suggesting that the level of conditional conservatism is less important in hightechnology firms relative to low-technology firms. Factors inducing conditional conservatism in high- and low-technology firms To test the effect of contracting (Hypothesis 1a), litigation risk (Hypothesis 1b), and taxation (Hypothesis 1c) on conditional conservatism in the high- and low-technology sector, we incorporate a proxy for each economic determinant in interaction with the Basu model. We estimate (3) to test the impact of each of these three factors as follows: X it / P it-1 = β 0 + β 1 D it + β 2 R it + β 3 D it * R it + β 4 DET it + β 5 DET it * D it + β 6 DET it * R it + β 7 DET it *D it * R it + ε it (5) Where X/P is earnings per share scaled by the beginning of the fiscal year price; R is the return on each firm from 9 months before the fiscal year-end to three months after the fiscal year-end; D is a dummy variable equal to 1 if returns are negative, and 0 otherwise; DET represents each of these three determinants of conditional conservatism: leverage, litigation risk and taxation. In (5), significant positive (negative) values of β 7 mean that greater values of the identified factor lead to higher (lower) levels of conditional conservatism. Hypothesis 1a investigates the positive relationship between the low level of leverage in high-technology firms and their low conditional conservatism degree. We test the hypotheses by the following model: X it / P it-1 = β 0 + β 1 D it + β 2 R it + β 3 D it * R it + β 4 LEV it + β 5 LEV it * D it + β 6 LEV it * R it + β 7 LEV it *D it * R it + ε it (5.1) Where X/P, R, and D as defined before; LEV represents a dummy variable equal to 1 if leverage [(Compustat #9 + Compustat #34) / Compustat #6)] is higher than the median sample, and zero otherwise. We run this model separately for each technological sector. High values of leverage are expected to induce mainly conditional conservatism in high- and lowtechnology firms. To study the effect of litigation risk on conditional conservatism, we run the following regression: X it / P it-1 = β 0 + β 1 D it + β 2 R it + β 3 D it * R it + β 4 LIT it + β 5 LIT it * D it + β 6 LIT it * R it + β 7 LIT it *D it * R it + ε it (5.2) 13

14 Where X/P, R, and D as defined before; LIT is a proxy for auditor litigation risk as used by Qiang (2007). It is a dummy variable that equals 1 if the code of a firm's auditor (Compustat #149) is from 1 to 8, but otherwise 0. Regarding (5.2), we focus only on the coefficient β 7, which shows how auditor litigation risk affects the asymmetric recognition of gains versus losses. In hypothesis 1b, we suppose that the demand of conditional conservatism is positively associated with the auditor litigation risk level. Thus, the coefficient β 7 is expected to be significantly positive for the high- and low-technology sample. Finally, to test the impact of taxation on conditional conservatism, we estimate the following regression separately for each sector: X it / P it-1 = β 0 + β 1 D it + β 2 R it + β 3 D it * R it + β 4 TAX it + β 5 TAX it * D it + β 6 TAX it * R it + β 7 TAX it *D it * R it + ε it (5.3) Where X/P, R, and D as defined before; TAX is a proxy for tax reduction cost reduction. Following Qiang (2007), the tax cost is the association between book income and tax income, β jt estimated from a time-series regression TX jt = β 0j + β jt BKTX jt + Ɛ jt for firm j over the sample period, where BKTX jt is tax expense for firm j in year t (Compustat # 16) and TX jt is tax expense minus deferred tax expense (Compustat # 16 - Compustat # 50); all variables are deflated by lagged total assets (Compustat # 6). The tax cost also generates incentives for conditional conservatism to reduce tax liabilities to the extent that taxable income and book income are associated. According to Hypothesis 1c, the sign of coefficient β 7 can be no significant. 5. Sample selection and empirical results Our original sample is drawn from the intersection of data from Compustat and CRSP for the period of 1974 to All data definitions are presented in the appendix. We then impose the following selection criteria: first, we delete firm-years with missing data for the research variables used in the regressions, and firm-years with negative total assets or book values of equity. Second, we divide our sample into high- and low-technology firms as defined by the three-digit U.S. Standard Industrial Classifications (SIC) codes used by Francis and Schipper (1999). According to their classification, we define high-technology industries as those operating in computer, electronics, pharmaceutical, and telecommunications industries. The high-technology sample consists of 16,396 firm-years. We also refer to the same classification of Francis and Schipper (1999) to define the low-technology sample and obtain

15 observations. Table1 presents a description of the sample classification into high- and lowtechnology firms. [Insert Table 1 here] In order to reduce the potential effect of extreme observations, all the continuous variables are winsorized at 1% and 99%. Table 2 summarizes descriptive statistics for key variables used in our regressions. The mean, standard deviation, median, first and third quartiles, and the minimum and maximum numbers are reported. Earnings-Price ratio is calculated as earnings per share scaled by price per share. For the high- (low-) technology sample, the mean of Earnings-Price ratio is (0.066). Returns are defined as the annual stock returns of the firm, measured compounding twelve monthly CRSP stock returns from nine months before the fiscal-year end to three months after the fiscal year-end. Both high- and low-technology firms are profitable and have positive returns, because the mean of returns in the high- (low-) technology sample is (0.178) and the median is (0.071). Size is the natural log of the market value of equity. The mean (median) of size is (4.975) for the high-technology firms, compared to (4.715). The MTB ratio is calculated as the market value of equity scaled by common equity. The average high- (low-) technology sample has a market-to-book ratio of (1.892). Leverage is the sum of short-term debt and long-term debt divided by total assets. The mean (median) of leverage is (0.084) for high-technology firms, compared to (0.248) for the lowtechnology sample. Then, similar to Chandra (2011), results show that the high-technology sample is less levered than low-technology one. [Insert Table 2 here] Auditor litigation risk is defined as Qiang (2007) by a dummy variable that equals 1 if the company s auditor is a big auditing firm. The mean and the median of auditor litigation risk for both samples are similar to those reported in prior literature (e.g., Qiang, 2007). Finally, the tax is calculated following Qiang (2007) by the association between book income and tax income. Mean (median) Tax is (0.898) for the high-technology sample, and (0.898) for the low-technology sample. These results are comparable to those reported by Qiang (2007). Table 3 displays the variables correlation matrix for our sample of 22,515 firm-years observations from 1974 to The upper (lower) right (left) portion reports the Pearson 15

16 (Spearman) correlations. The variables that have a positive relation to Earnings-Price ratio, both in Pearson and Spearman, are return, Size, Lev, and, LIT, which are consistent with Khan and Watts (2009). However, the Earnings-Price ratio is significantly negatively correlated with MB. The positive correlation between Earnings-Price ratio and return indicates that accounting earnings capture the information reflected in returns. [Insert Table 3 here] Conditional conservatism in U.S. firms Table 4, Panel A, reports the level of conditional conservatism in the full sample using the Basu model. The coefficient of D*R is positive and significant as predicted (β 3 = 0.156, P- value < 0.01), suggesting that U.S. firms recognize bad news in a timely fashion than good news. This indicates that American firms are conservative on average. Table 4, Panel B, presents the results from estimating the mean values of C_Score and G_Score. Consistent with our expectations, C_Score (G_Score) has a mean of (0.072) and a median of (0.063), indicating that the average of C-Score, which reflects the incremental timeliness of bad news is higher than the average of G_Score, which reflects the timeliness of good news. This means that our sample is conservative on average, which is consistent with our findings using the Basu model and the results reported by Khan and Watts (2009). [Insert Table 4 here] Conditional conservatism in high- and low-technology Hypothesis 1 predicts that conditional conservatism in U.S. high-technology firms is lower compared to low-technology firms. Panel A of Table 5 reports the results based on time-series averages of annual cross sectional coefficient estimates from (1). Consistent with our expectations, the average yearly value of the asymmetric timeliness coefficient β 3 is for the high-technology sample, which is lower than for the low-technology sample. However, this difference is not significant. Moreover, our results show that the average yearly value of the Basu model Intercept is lower for the high-technology (0.038) sample than for the low-technology one (0.076), and this difference is significant (P-value < ), which suggests that the level of unconditional conservatism is significantly higher in hightechnology firms relative to low-technology firms. Overall, our findings are consistent with those of Chandra (2011) that the level of conditional conservatism is less important in hightechnology firms compared to low-technology firms although this difference is not 16

17 significant; that high-technology firms are more unconditionally conservative than lowtechnology firms. Panel B of Table 5 compares the mean values of C_SCORE between high- and lowtechnology samples. Low-technology firms have a higher mean of C_SCORE (0.500) than high-technology firms (0.089), and this difference is highly significant (P-value < ). Our results based on C_Score are consistent with the level of conditional conservatism, which is lower in high-technology firms than low-technology firms. In summary, results in Table 5 indicate that conditional conservatism is less important in high-technology firms than lowtechnology firms. [Insert Table 5 here] Furthermore, using the results of Table 5, we plot the yearly Basu coefficient and the Basu model Intercept. The results are shown in Figure 1 and Figure 2. [Insert Figure 1 here] [Insert Figure 2 here] Figure 1 shows a time-series plot of the median Basu coefficient. It indicates that the asymmetric timeliness coefficient (a measure of conditional conservatism) is lower for the high-technology sample in comparison to the low-technology sample in 22 of 37 years. It ranges from to in high-technology firms and from to in lowtechnology firms. Consequently, the level of conditional conservatism is lower for the hightechnology sample compared to the low-technology sample. Figure 2 shows the changing of the Basu model intercept (a measure of unconditional conservatism) across years. It indicates that the intercept is lower for high-technology firms than for low-technology firms in 30 of 37 years. It ranges from to in the hightechnology sample and from to in the low-technology sample. Consequently, the level of unconditional conservatism is higher for high-technology firms than for lowtechnology firms. Panel A (B) of Table 6 reports the generalized linear regression results for a pooled sample of (6.036) firm-year observations over the period of 1974 to 2010 for the high- (low-) technology sample. The first column of panel A (B) reports the Basu coefficient for the high (low) technology sample. The others include the factors inducing conditional conservatism in high- (low-) technology firms. 17

18 Results show that the coefficient on returns (R) is significantly negative for the hightechnology firms (β 2 = 0.092, P-value < 0.01), indicating that good news are negatively associated to accounting earnings. However, for the low-technology sample, the coefficient on returns is positive and significant (β 2 = 0.027, P-value < 0.01), which is consistent with US previous studies (Basu, 1997; Kwon et al., 2006; LaFond and Watts, 2008; Khan and Watts, 2009). Further, we find that in both sub-samples, the coefficient of D*R is positive and significant. This indicates that high- and low-technology firms are conservative. That is, on average, the association between earnings and bad news is stronger than between earnings and good news. Consistent with Hypothesis 1, which predicts that conditional conservatism in U.S. hightechnology firms is lower compared to low-technology ones, the results document that the asymmetric timeliness coefficient is lower in the high-technology sample (β 3 = 0.092, P-value < 0.01) than in low-technology one (β 3 = 0.172, P-value < 0.01). This means that the level of conditional conservatism in high-technology firms is less important than in low-technology firms. In other words, high-technology firms are less conditionally conservative. In addition, the results reported in the first column of Table 6 show that the model intercept, β 0 which measures the unconditional conservatism, is lower in the high-technology sample than in the low-technology sample. This evidence suggests that high-technology firms exhibit more unconditional conservatism than low-technology firms. [Insert Table 6 here] The drivers of the conditional conservatism in high- and low-technology firms Column 2 of Table 6 reports the estimation results for Model (3.1). This model examines how the level of leverage affects the asymmetric timeliness of recognition of bad news vs. good news for the high-technology and low-technology firms separately. The coefficients of the two-way interaction between D*R (β 3 ) are still positive and significant for both hightechnology firms (0.082, P-value < 0.01) and low-technology firms (0.073, P-value < 0.01). These results indicate that less levered high- and low-technology firms are conditionally conservative. Regarding the three-way interaction term, LEV*D*R, the coefficient reflects the incremental effect of leverage on conditional conservatism. Results show that the coefficients are significantly positive for the high-technology sample (β 7 = 0.059, P-value < 0.01) and the low-technology sample (β 7 = 0.076, P-value < 0.01), which is consistent with the idea that earnings exhibit greater conditional conservatism when the level of leverage increases 18

19 (LaFond and Watts, 2008; Khan and Watts, 2009). In other words, more levered firms are more conservative than less levered firms. These results are consistent with the existence of debt-contracting-based demands for conditional conservatism. Basu (1997), Ball and Shivakumar (2005), and Ball et al. (2008) argue that the most widely accepted source of conditional conservatism is shaped primarily by debt contracts. In fact, low-technology firms have more tangible assets and are more levered than high-technology firms. This suggests a higher contracting demand for conditional conservatism from low-technology firms and a lower demand from high-technology firms due to their lower level of leverage. The combination of these results suggests that the level of leverage is an important determinant of conditional conservatism and that firms with higher leverage report more conservative earnings regardless of their technology sector. Overall, in the first and second column of Table 6, our results provide evidence that both high-technology and low-technology firms exhibit conditional conservatism. Moreover, hightechnology firms are less conditionally conservative than low-technology firms and the degree of conditional conservatism increases with the level of leverage irrespective of their technology level. Regarding auditor litigation risk driven motivations for conditional conservatism, results reported in the third column of Table 6 shows that the LIT*D*R interaction term is insignificant for the high-technology sample (-0.011), implying that the degree of auditor litigation risk does not alter the asymmetric timeliness of recognition of bad news vs. good news in high-technology firms. However, the coefficient on the LIT*D*R interaction term is marginally significant for the low-technology sample (0.309, P-value < 0.10), which provides evidence that a greater degree of auditor litigation risk seems to induce more conditional conservatism in low-technology firms. This is probably because low-technology firms have more tangible assets, so the risk of overvaluation of net assets is higher. Consequently, the auditor litigation risk will be higher since firms are more likely to be sued when they overstate their assets than understate them (Kellog, 1984; St Pierre and Anderson, 1984). That is why auditors are more motivated to be more conservative. However, regarding high-technology firms, it is possible that the level of auditor litigation risk does not affect the earnings timeliness, because they have fewer tangible assets and, the risk of litigation is lower. Overall, the results provide evidence consistent with low-technology firms incorporate bas news as losses in a timelier manner than good news as gains as the level of auditor litigation risk increases. In the case of high-technology firms, a greater auditor litigation risk does not drive 19

20 a demand for conditional conservatism. The combined results support our third hypotheses and indicate that the demand for conditional conservatism is driven by the level of auditor litigation risk in low-technology firms and that the level of auditor litigation risk is unlikely to have an impact on conservatism in high-technology firms. Thus, in order to reduce litigation costs, only low-technology firms demand more conditional conservatism. Hypothesis (4) predicts that the demand of high-technology firms for conditional conservatism is not related to their taxable earnings. Column 3 of Table 6 reports the coefficient estimates and p-values. Panel A presents results for the high-technology sample and panel B reports results for the low-technology sample. Taxation does not induce conditional conservatism, which is consistent with the coefficients on the three-way interaction between TAX*D*R and are insignificant in both high-technology (-0.004, P-value > 0.10) and low-technology firms (0.020, P-value > 0.10). This shows that conditional conservatism is unlikely to be desirable in tax planning. It is possible that taxation costs do not motivate firms to adopt conditional conservatism because as founded by Qiang (2007), taxation induces only unconditional conservatism. She explained that conditional conservatism is uncontrollable, introduces volatility, and is, not neutral but undesirable in tax planning. Overall, the results reported in column 3 lend support to the fourth hypothesis and indicate that taxation does not alter the asymmetric timeliness of recognition of bad news vs. good news regardless the technology level of firms. After investigating the impact of each determinant on conditional conservatism, we repeat tests including the previous determinants in the same model. All these three factors are added to the Basu's model and interacted with each of the variables for the high-technology and lowtechnology samples separately. The results are reported in the last column of Table 6. For high-technology firms, the results are similar to the main results reported for each determinant. This indicates that the leverage induces conditional conservatism in hightechnology firms (0.052, P-value < 0.01), but auditor litigation risk (0.011, P-value > 0.10) and taxation (0.0008, P-value > 0.10) does not drive conditional conservatism. In case of lowtechnology firms, we same results only for leverage (0.066, P-value < 0.01) and auditor litigation risk (0.379, P-value < 0.01), suggesting that both leverage and auditor litigation risk create a demand for conditional conservatism in low-technology firms. However, the results in the full model show that taxation has also a marginal positive effect on conditional conservatism in the low-technology sample (0.026, P-value < 0.10). 20