Is the accrual anomaly robust to firm-level analysis?

International Review of Financial Analysis 34 (2014) 157–165

Contents lists available at ScienceDirect

International Review of Financial Analysis

Is the accrual anomaly robust to firm-level analysis? Maria Strydom a,⁎, Michael Skully a, Madhu Veeraraghavan b a b

Monash University, Australia T.A. Pai Management Institute, India

a r t i c l e

i n f o

Article history: Received 23 April 2014 Accepted 17 June 2014 Available online 25 June 2014 JEL classification: M41 G12 G14 G38

a b s t r a c t This study investigates whether firm-level accrual mispricing exists and if such mispricing is persistent. Our results show both under and overpricing of accruals that persevere. Specifically, we show that a trading strategy going a dollar long (short) in underpriced (overpriced) accrual firms yields significant abnormal returns in most years investigated. We examine whether firm characteristics such as size, analyst following and real activities management can explain why some firms are mispriced and others not. Our findings show that firm-level mispricing differs from that documented at the country-level. Whilst the country-level anomaly seems to have diminished; the firm-level accrual anomaly remains. © 2014 Elsevier Inc. All rights reserved.

Keywords: Accrual anomaly Mispricing Firm-level

1. Introduction and motivation This study examines the firm-level mispricing of accruals. The accrual anomaly suggests that investors overestimate the persistence of the accrual component of earnings and subsequently misprice it (Sloan, 1996). Abnormal returns are available to a strategy of buying (selling) low (high) accrual firms, and so the existence of the anomaly is seemingly evidence against market efficiency. The anomaly is limited to certain subsets of firms (profit-making firms, Dopuch, Seethamraju, & Xu, 2010; low disclosure quality firms, Drake, Myers, & Myers, 2009; and smaller firms in high sentiment periods, Ali & Gurun, 2009). Dopuch et al. (2010) document mispricing of profit-making firms' positive accruals but conclude that loss firms are accurately priced. When low disclosure quality firms are excluded from consideration, mispricing is substantially reduced (Drake et al., 2009). Ali and Gurun (2009) show that investors misprice small firms particularly during high sentiment periods. These studies therefore suggest that specific firm characteristics such as being a profit firm, having low disclosure quality or being small are associated with mispriced accruals. The accrual anomaly therefore appears to be driven by firm-level characteristics and whilst Ali and Gurun (2009), Drake et al. (2009) and Dopuch et al. (2010) examine subsets of accrual anomaly firms, they do not investigate individual firm-level accrual mispricing and its persistence. ⁎ Corresponding author at: Monash University, Caulfield Campus. 900 Dandenong Rd, Caulfield East, 3145, VIC, Australia. Tel.: +61 3 99034581. E-mail address: [email protected] (M. Strydom).

http://dx.doi.org/10.1016/j.irfa.2014.06.001 1057-5219/© 2014 Elsevier Inc. All rights reserved.

Fama (1998) conjectures that efficient markets create events which suggest that prices overreact to information. However, he concludes that in an efficient market, overreactions to information should be about as frequent as underreactions. Therefore if we find at the firmlevel that investors overestimate the persistence of accruals about as often as they underestimate it, the existence of such mispricing is not necessarily inconsistent with efficient market theory. We are therefore motivated to examine firm-level over and underpricing of accruals to determine its prevalence and persistence and shed some light on the true market efficiency implications of the accrual anomaly. We are the first study to examine the firm-level accrual anomaly. This paper therefore has two main objectives. The first is to determine whether firm-level mispricing exists and is persistent. We estimate firm-level accrual mispricing variables and examine whether these firms remain mispriced over time. We are particularly interested to determine whether firms have over or underpriced accruals, and how long these remain so. The second objective is to examine the characteristics of mispriced firms. That is, we estimate whether firms with mispriced accruals have common characteristics. Several studies are directly related to ours. Ali and Gurun (2009) show that the country-level anomaly is most common in subsets of small firms during high sentiment periods. Drake et al. (2009) document that the anomaly is present for low disclosure quality firms only whilst Dopuch et al. (2010) confirm that it exists solely for profitmaking firms. Whilst these studies show that the country-level accrual anomaly exists only for certain subsets of firms, our study instead investigates the pricing of accruals at the firm-level and documents the

158

M. Strydom et al. / International Review of Financial Analysis 34 (2014) 157–165

existence of under and overpriced firms. Green, Hand, and Soliman (2011) document that returns to accrual anomaly based hedge fund trading have decreased post-SOX. This study goes further by showing that whilst firm-level mispricing persistence similarly decreased postSOX it has since increased again and remains fairly consistent longterm. We conclude by examining whether firms identified as mispriced have common characteristics. Our results indicate that some 13% of firms have mispriced accruals in any given year. This mispricing appears pervasive with 83% of significantly over- and underpriced firms remaining so for at least one-year, 70% for at least two years, and 51% for more than four years. Furthermore, a trading strategy of buying underpriced firms and shorting overpriced firms over the 12-year sample period would have almost doubled one's investment (return of 93%). Additional analysis also reveals that firm-level mispricing has remained fairly constant over time. The percentage of firms with underpriced accruals, however, seems to have increased over the sample period whilst those with overpriced accruals have decreased. Mispricing persistence has increased in the latter part of the 2000s after decreasing in earlier years. This suggests that the decrease in the country-level accrual anomaly documented (Cohen & Zarowin, 2010; Green et al., 2011) may have been temporary. Studies document post-SOX improvements in earnings quality resulting in decreased discretionary accruals (Iliev, 2010; Singer & You, 2011) thus suggesting that the anomaly may have been attenuated. We show instead that mispriced firms have significantly higher real activity based earnings management post-SOX and that firm-level accrual mispricing persists. Mispriced firms are found to have larger total asset values and increases in analysts' coverage of these firms appear to be associated with a reduction in the pervasiveness of mispricing. Our study contributes to the accrual anomaly literature in several ways. First, we show that some firms have overpriced accruals, others have underpriced accruals whilst yet others still are accurately priced. We therefore illustrate the importance of investigating accrual mispricing at the firm-level. Our second contribution is documenting the persistence of firm-level accrual mispricing and illustrating that a strategy of buying underpriced accrual firms and selling overpriced accrual firms yields significant trading profits. The final contribution of this study is confirming that the absolute level of firm accruals has decreased post-SOX. However, whilst prior studies suggest that this is due to decreased earnings management we show instead that mispriced firms are significantly more likely to manage accruals through real activity based earnings management techniques postSOX. Thus, SOX did not necessarily decrease earnings management; rather it just resulted in a change from accrual-based to real activity based earnings management. Investors, in turn, seem equally unable to accurately price the lower quality earnings stemming from real activity based earnings management than from accrual-based earnings management. The rest of this study is structured as follows: The next section presents the relevant literature and develops testable hypotheses. Data and methodology are discussed in Section 3 and empirical findings in Section 4. Section 5 concludes.

suggest that the anomaly is present only for certain subsets of firms (Ali & Gurun, 2009; Dopuch et al., 2010; Drake et al., 2009). The accruals of small firms, for instance, are potentially mispriced given that they are typically followed mainly by individual investors (as opposed to more sophisticated institutional investors or analysts) (Ali & Gurun, 2009). Drake et al. (2009) propose that investors faced with low quality financial disclosures are more likely to misprice accruals and document reduced accrual mispricing for higher disclosure quality firms. Dopuch et al. (2010), in turn, propose that earnings for loss firms are less value relevant and that they are therefore less likely to be mispriced in the accrual anomaly. Their results confirm this and they document that it is the positive accruals of profit firms that are mispriced and that loss firms do not have mispriced accruals. Taken together, these studies suggest that Sloan's (1996) countrylevel accrual anomaly relates only to certain subsets of underlying firms and suggest that there are certain types of firms more likely to have mispriced accruals than others. They do not, however, investigate the mispricing of accruals at the firm-level. In fact, no study to date has done so. We expect, based on the extant literature and in particular Fama (1998) that some firms will be overpriced, others underpriced and still others have no accrual mispricing at all.2 This discussion leads to our first hypothesis: Hypothesis 1. Accruals are mispriced at the firm-level. More recently, studies show a decrease in the profitability of the country-level accrual anomaly based trading strategy (Bhojraj & Swaminathan, 2009; Green et al., 2011; Keskek et al., 2013; Richardson, Tuna, & Wysocki, 2010). Whilst these studies credit stricter regulation and better disclosure quality for this reduction, we suggest that this might not be so. If more stringent regulation truly resulted in the country-level accrual anomaly's demise, the underlying firm-level mispricing of accruals should also be mitigated or significantly reduced (more accurate financial disclosures should allow better pricing decisions of individual firm accruals much like at the country-level). Alternatively, if firm-level mispricing remains, then the lower countrylevel mispricing documented in the late 2000s could simply be an “averaging” effect of underlying firm over and underpricing. It is our expectation that whilst some firms may have overpriced accruals (as in the country-level accrual anomaly) others may be underpriced whilst yet others may be accurately priced. So, if there are equal numbers of under and overpriced accrual firms, a countrylevel accrual anomaly might not be documented at a particular point in time even when a large proportion of underlying firms are mispriced (as the over and underpricing might cancel each other out) consistent with Fama (1998). Whether underlying firm-level accrual mispricing therefore persists over time or rather corrects due to investors recognizing its implications remains unknown. That is, we examine whether (and how long) firm-level accrual mispricing persists. This discussion leads to our second hypothesis: Hypothesis 2. Firm-level accrual mispricing persists over time.

3. Data and methodology

2. Background and hypotheses development

3.1. Data

The accrual anomaly literature documents abnormal returns to a strategy of selling high accrual firms and buying low accrual firms (Sloan, 1996) on the premise that investors overestimate (underestimate) the persistence of the accrual (cash flow) component of earnings. Subsequent studies confirm the anomaly and its persistence at the country-level (Hirshleifer, Teoh, & Yu, 2011; Lev & Nissim, 2006; Mashruwala, Rajgopal, & Shevlin, 2006; Shi & Zhang, 2012).1 Others

The data required for calculating the accrual mispricing variables are obtained from the Compustat/Center for Research in Security Prices (CRSP) database. The variables for the mispricing tests include earnings (EARt) and its accrual (ACCt) and cash flow (CFOt) components, and sizeadjusted (abnormal) returns (Rt + 1). The calculation of each of these

1 More recently though it appears the country-level anomaly is diminished or completely mitigated (Green et al., 2011).

2 Fama (1998) suggests that if over and underreactions to information are about equal in their frequency, then this is consistent with market efficiency and so no anomaly really exists. We are therefore also interested in the investigating the frequency of over and underpricing of accruals.

M. Strydom et al. / International Review of Financial Analysis 34 (2014) 157–165

variables is consistent with that in Kraft, Leone, and Wasley (2007) and Sloan (1996) and is discussed below. Earnings are measured as current period earnings (Compustat item #178) scaled by average total assets (Compustat item #6). Accruals are estimated per the balance sheet approach as follows: ACC i;t ¼ ðΔ CA−Δ CASH Þ− ðΔ CL−Δ STD−Δ TPÞ−DEP t

ð1Þ

where ACCt is current period accruals; ΔCA is change in current assets (change in Compustat item #4 from periodt − 1 to periodt); ΔCASH is the change in cash & cash equivalents (change in Compustat item #1 from periodt − 1 to periodt); ΔCL is the change in current liabilities (change in Compustat item #5 from periodt − 1 to periodt); ΔSTD is the change in debt included in current liabilities (change in Compustat item #34 for periodt − 1 to periodt); ΔTP is the change in income tax payable (change in Compustat item #71 from periodt − 1 to periodt); and DEPt; is depreciation and amortization expenses (Compustat item #14). The calculated accrual value is scaled by average total assets (Compustat item #6). Cash flow is measured as the difference between the calculated earnings and accrual values: CFOi;t ¼ EARi;t −ACC i;t

ð2Þ

where CFOi,t is the cash flow component or earnings, EARi,t is current year earnings (Compustat item #178), and ACCi,t is current period accruals as per Eq. (1). Each of these variables is scaled by average total assets (Compustat item #6). Size-adjusted buy and hold returns (Rt + 1) are calculated as the raw 12-month buy and hold return for a period starting 4 months after a firm's fiscal year end adjusted for the return on the size-matched decile portfolio over the same period. Additional firm variables were obtained to analyze the characteristics of significantly mispriced firms, including size and sector data, exchange listings, and analyst following. The size and sector data was acquired from Compustat. Size was measured as the dollar value of total assets (in millions as in Compustat item #6). Sector data was obtained for all NYSE NYSE/NASDAQ and AMEX firms in the Compustat universe. All firms were classified as either energy, materials, industrials, consumer discretionary, consumer staples, healthcare, information technology, telecommunications, or utilities, based on their twodigit Global Industry Classification System (GICS) code. Exchange listing information was sourced from the merged Compustat/CRSP database, whilst analyst following data was retrieved from the Institutional Brokers' Estimate System (I/B/E/S) database. All analyst recommendations for sample firms are from the I/B/E/S. Analyst following is estimated as the number of recommendations made for a firm in each year. Where a firm was not followed by analysts (and so had no recommendations), analyst following was equal to zero in that year. Real earnings management was estimated with two comprehensive measures (abnormal production and abnormal discretionary expenses) as in Roychowdhury (2006) and data for this was obtained from the Compustat database.3 3.2. Sample selection The initial sample consisted of all NYSE, NASDAQ and AMEX listed companies on the merged Compustat/CRSP database for the period 1965–2011 (222,582 firm-year observations for 15,106 firms) (see Table 1). Of these, 37,164 firm-year observations (2,878 firms) were for firms in the financial sector (GICS sector code 40) and were

3

Additional data items required to estimate the Roychowdhury (2006) real earnings management proxies included sales (compustat item # 12), cash flow from operations (compustat item # 308–#124), production costs (compustat item #41 + #3) and discretionary expenses (compustat items # 45+ #46+ #189).

159

Table 1 Sample selection. Firm-years Compustat/CRSP database observations for the sample period 1965–2011 for firms listed on NYSE, NASDAQ and AMEX Less: GICS financial firms Less: missing earnings or total asset data Less: missing CRSP monthly returns data Less: extreme observations (1% trimming) Less: firms with less than 25 yrs of observations in the sample Final sample

Number of firms

222,582

15,106

(37,164) 185,418 (20,963) 164,455 (1,117) 128,452 (4,590) 123,862 (59,432)

(2,878) 12,228 (1,017) 11,211 (267) 9,268 (312) 8,956 (7,036)

64,430

1,920

This table reports on the selection of the sample (including data from 1965 to 2011) for this study. Only firms for which at least 25 years of data (in the sample period) is available are included in this study as regressions estimated with very small sizes (n) often yield biased estimates. The requirement for at least 25 years of observations per firm is required for both hypotheses. Whilst for the first hypothesis any period of 25 years within the sample period of 1965–2011 suffices for inclusion, the firm-specific regressions for Hypothesis 2 require that the 25-year data requirement be met for each estimation period. That is, for a firm to be included for 1995 it requires data for 25 years within the period 1965–1994. Further information on the exact number of firm-year accrual mispricing variables estimated for each year is available from Table 4.

eliminated, leaving 185,418 firm-year observations for 12,228 firms.4 In addition, earnings or total asset data was missing for 20,963 firmyears, and deletion of these left 164,455 firm-year observations for 11,211 firms. A further 1,117 firm-year observations lack monthly returns data from the CRSP database and exclusion of these reduced the sample to 128,452 firm-year observations (9,268 firms). Consistent with prior research (De Fond & Park, 2001; Kraft, Leone, & Wasley, 2006), the top and bottom 1% of values were trimmed to eliminate the effects of extreme observations further reducing the sample to 123,862 firm-year observations. For Hypothesis 1, we require at least 25 firm-year observations for each firm over the sample period 1965–2011 (that is 25 years of data). For Hypothesis 2 twenty-five consecutive years of observations are required for each firm in each period where a firm-year mispricing variable is estimated.5 This requirement is set to limit the potential errors from including too few observations in a regression. With at least 25 observations for each firm in a standard OLS model with only two independent variables, biased estimates or large error terms should be mostly eliminated. For Hypothesis 1 this process eliminates 59,432 firm-year observations for 7036 firms (leaving 64,430 observations for 7036 firms) whilst the final sample for the estimation of firm-year mispricing variables is evident from Table 4. 3.3. Methodology 3.3.1. Firm-level accrual mispricing: the Kraft et al. (2007) model Estimating accrual mispricing requires determination of how investors perceive the contribution of accruals to earnings persistence compared to what its actual contribution is. Sloan (1996) employs a rational pricing test of accrual numbers from Mishkin (1983) which has been used in most accrual mispricing studies to date (Collins & Hribar, 2000; Fairfield, Whisenant, & Yohn, 2003; Rangan & Sloan, 1998). Mishkin (1983) specifies, however, that the model is best suited to very large samples and will likely yield biased results for small 4 Financial firms were excluded due to difficulties in calculating accruals and their different nature (Beneish & Vargus, 2002; Desai, Rajgopal, & Venkatachalam, 2004; Lev & Nissim, 2006; Mashruwala et al., 2006). 5 For example for 1995, we require that a firm have at least 25 years of data in the period 1965–1994.

160

M. Strydom et al. / International Review of Financial Analysis 34 (2014) 157–165

Table 2 Descriptive statistics for the sample. Variable

N

Mean

Median

Standard deviation

Minimum

Maximum

Skewness

Kurtosis

ACCt CFOt EARt + 1 Rt + 1

64,430 64,430 64,430 64,430

−0.028 0.128 0.100 −0.128

−0.032 0.132 0.099 −0.108

0.088 0.128 0.115 0.711

−0.503 −1.100 −1.148 −6.927

0.737 0.655 0.565 3.170

1.066 −1.547 −1.938 −2.74

9.166 10.034 14.825 20.93

This table presents the descriptive statistics for the sample employed to determine whether firm-level accrual mispricing exists and employs data from 1965 to 2011. Here ACCt is current period accruals calculated as per the balance sheet method, ACCt = (ΔCA − ΔCASH) − (ΔCL − ΔSTD − ΔTP) − DEPt , where ΔCA is change in current assets (Compustat item #4), ΔCASH is change in cash/cash equivalents (Compustat item #1), ΔCL is change in current liabilities (Compustat item #5), ΔSTD is change in debt included in current liabilities (Compustat item #34), ΔTP is change in income tax payable (Compustat item #71), and DEPt is depreciation and amortization expenses (Compustat item #14). The accrual value calculated is scaled by total assets (Compustat item #6), CFOt is current year operating cash flow calculated as EARt − ACCt and scaled by total assets (Compustat item #6), EARt + 1 is one year ahead earnings (Compustat item #178) scaled by total assets (Compustat item #6), Rt + 1 is size-adjusted buy and hold returns to a security for a 12-month period starting 4 months after the firm's fiscal year end as in Sloan (1996).

samples. The Mishkin model is therefore not suitable for estimating firm-level accrual mispricing that will necessarily have a small sample (given the nature and availability of data). Kraft et al. (2007) suggest an OLS model alternative which allows for inclusion of additional variables and is shown to yield similar results to the Mishkin model. The OLS model can also be applied to smaller samples as it involves simple least squares estimation. This model is therefore employed to estimate firm-level mispricing in this paper. It is estimated as: Ri; tþ1 ¼ α 0 þ α 1 ACC i;t þ α 2 CFOi;t þ εi;tþ1 :

ð3Þ

Ri,t + 1 is one period ahead size-adjusted returns estimated tear on year starting for months after the end of the financial year. ACCi,t is accruals estimated as per the balance sheet method, CFOi,t is the cash flow component of earnings, estimated as the differences between earnings and accruals. εi,t + 1 is the stochastic error term from the regression. In Eq. (3), α0 is the intercept term, whilst α1 is the coefficient of accruals which, if significant, indicates the existence of accrual mispricing. A negative sign on the accrual coefficient shows overpriced accruals whilst a positive one indicates underpriced accruals. The coefficient of cash flow, α2, similarly measures whether cash flows are accurately priced. 4. Results and discussion 4.1. Firm-level mispricing The mean (median) accrual value for the sample is − 0.028 (− 0.032), whilst the standard deviation is 0.088 (see Table 2). The mean (median) value of earnings is 0.10 (0.099), whilst that of cash flow is 0.128 (0.132). The distribution of the earnings and cash flow values around the mean is similar, with standard deviations (kurtosis) of 0.115 (14.83) and 0.128 (10.03), respectively. Size-adjusted buy and hold returns have a mean (median) value of − 0.128 (− 0.108), indicating that returns are, on average, negative in the sample period. The accrual mispricing values obtained from firm-level estimations of the Mishkin model (see Table 3) reveal that, of the 1,913 firm-level anomalies calculated over the sample period, 918 firms have overpriced accruals, of which 92 are significant.6 Of the 994 firms with underpriced accruals, 149 are significant. Thus, 241 out of 1,913 firms, or 12.6% of sample firms, are significantly over (4.8%) or underpriced (7.8%) in the period investigated. It is interesting to note that there are approximately even numbers of over (918) and underpriced (994) accruals firms in the sample. As we document significant overpricing and underpricing of accruals at the firm-level, our results support Hypothesis 1, which predicts

6 The majority of firms (N70%) with significant mispricing are mispriced at the 1% or 5% level. However, as 10% significance level is commonly employed in studies of this nature we report results as such.

that firm-level mispricing exists and we show that it differs from that at the country-level. In terms of examining the persistence of firm-level accrual mispricing (that is whether firms mispriced in one-year remains so in subsequent periods) we estimate firm-level accrual mispricing variables for the period 1995–2011. A rolling 30-year period is employed for each firm to estimate one firm-year mispricing variable (i.e. the mispricing variable for 1995 is estimated for each firm with data from 1965–1994).7 The results, presented in Table 4, show not only the percentage of firms that are significantly under and overpriced but also the persistence of such mispricing in subsequent years. Panel B of Table 4 reveals that on average, 13% of all firms are mispriced in any given year. Of these 57% are underpriced and 43% are overpriced. At the firm-level, there are therefore slightly more underpriced firms than overpriced ones.8 On average, some 83% of significantly mispriced remain so for at least one-year, 70% for at least two and 51% for four or more years. It therefore appears as if firm-level accrual mispricing is persistent and slow to be corrected. Whilst the one-year persistence remains fairly consistent around 83%, more variation is visible when looking at two and four-year figures. The two-year persistence figures decrease from 75% for firms mispriced in 1999 to a low of 60% in 2001 and remain at those low levels until about 2005. By 2007 the two-year mispricing persistence has increased to the pre-2000 levels of around 75%. A similar decrease and subsequent increase in the four-year persistence of firm-level accrual mispricing are also visible. This is reconcilable with country-level accrual anomaly results that show a substantial decrease in the accrual anomaly in the post SarbanesOxley Act (2002) and Shi and Zhang (2012) period (Green et al., 2011). However, in contrast to those studies, we show that the persistence of mispricing increases again in the latter part of the 2000s and that the regulatory changes therefore do not appear to have had a lasting impact on reducing accrual mispricing. Next, to determine whether investors could profit from a mispriced accrual firm based trading strategy, a portfolio long $1 in underpriced firms and short $1 in overpriced firms are created.9 The abnormal returns (measured as the abnormal buy and hold returns for the year commencing the month after the end of the financial year in which the firm is identified as being mispriced) from investing in such a portfolio are presented in Table 5. It documents that $1 invested at the start of the 16-year sample almost doubles over the period. The abnormal returns are positive in most years. Firm-level mispricing is therefore

7 The requirement here is that there are at least 25 years of data available for the firm in the preceding 25-year period. 8 We are cautious in stating this is necessarily so as our sample period is limited. Whilst the number of overpriced firms has decreased over the sample period, the number of underpriced firms has increase leaving open the possibility that there is some pattern and that over a longer term equal numbers of over and underpriced firms will be documented. 9 This strategy includes buying significantly (at least at the 10% level) accrual firms and selling significantly overpriced accrual firms.

M. Strydom et al. / International Review of Financial Analysis 34 (2014) 157–165

161

Table 3 Analyses of firm-level accrual mispricing. n

Average accrual coefficient Average p-value % Significant at the 10% level % Significant at the 5% level % Significant at the 1% level

Significantly mispriced firms 241 (12.6%) −0.47

0.045⁎⁎

43.98%

37.3%

18.7%

Overpriced accrual firms All overpriced firms Significantly overpriced

918 (48%) 92 (4.8%)

−0.314

0.0456⁎⁎

46.7%

29.4%

23.9%

Underpriced accrual firms All underpriced firms Significantly underpriced Total firms

994 (52%) 149 (7.8%) 1,913

10.528

0.0449⁎⁎

42.3%

42.3%

15.4%

This table presents results from the estimation of firm-level accrual mispricing values for the period 1965–2011. It specifies the number of over and underpriced firms in the sample and provides the average coefficient of accrual mispricing from the Kraft, Leone and Wasley model. The KLW (Kraft et al., 2007) model is estimated as: Ri;tþ1 ¼ ACC i;tþ CFOi;t þ ε i; t where Ri,t + 1 is one year ahead size-adjusted buy and hold returns to a security for a 12-month period starting 4 months after the firm's fiscal year end as in Sloan (1996). ACCt is the current period accruals calculated as per the balance sheet method, CFOt is the current year operating cash flow calculated as EARt − ACCt and scaled by total assets. One mispricing variable is calculated for each firm in the sample over the sample period, resulting in 6844 firm-level anomalies over the period 1965–2011. ⁎⁎ indicates significance at the 5% level.

not only persistent, but investors can also profit from trading on it. These findings support Hypothesis 2, that accrual mispricing persists over time. Indeed, approximately 51% of significantly mispriced accrual firms remain so for more than four years.

4.2. Characteristics of significantly mispriced firms The characteristics of significantly mispriced firms are presented in Tables 6 and 7. It compares (in Panel A of Table 6) the sector breakdown

Table 4 Firm-level accrual mispricing persistence. Year

n

Significant

Significantly underpriced accruals

Significantly overpriced accruals

Significant at 10% level

Significant at 5% level

Significant at 1% level

Mispricing persists for 1 year

Mispricing persists for 2 year

Mispricing persists for 4+ year

Panel A: firm-level pricing in numbers 1995 1,070 130 1996 1,085 129 1997 1,104 131 1998 1,153 141 1999 1,271 162 2000 1,281 170 2001 1,275 167 2002 1,256 172 2003 1,235 164 2004 1,207 153 2005 1,200 152 2006 1,152 139 2007 1,141 144 2008 1,157 151 2009 1,174 150 2010 1,251 160 2011 1,312 170 Average 152

55 70 69 69 90 92 87 97 102 97 94 89 91 92 90 95 94 87

75 59 62 72 72 78 80 71 62 56 58 50 53 59 60 65 76 65

90 92 57 64 61 70 75 74 69 57 56 44 54 67 61 65 70 66

30 26 52 55 72 70 63 61 67 67 60 61 60 52 58 61 66 58

10 11 22 22 29 30 29 33 28 29 36 34 30 32 31 34 34 58

114 102 110 118 142 139 134 129 127 123 117 117 127 132 124 141 . 125

90 89 97 105 121 115 101 106 102 98 100 106 113 114 110 . . 104

66 71 77 73 78 76 70 70 70 76 81 83 88 . . . . 75

Panel B: firm-level pricing in percentages 1995 1,070 12% 1996 1,085 12% 1997 1,104 12% 1998 1,153 12% 1999 1,271 13% 2000 1,281 13% 2001 1,275 13% 2002 1,256 14% 2003 1,235 13% 2004 1,207 13% 2005 1,200 13% 2006 1,152 12% 2007 1,141 13% 2008 1,157 13% 2009 1,174 13% 2010 1,251 13% 2011 1,312 13% Average 13%

42% 54% 53% 49% 56% 54% 52% 56% 62% 63% 62% 64% 63% 61% 60% 59% 55% 57%

58% 46% 47% 51% 44% 46% 48% 41% 38% 37% 38% 36% 37% 39% 40% 41% 45% 43%

69% 71% 44% 45% 38% 41% 45% 43% 42% 37% 37% 32% 38% 44% 41% 41% 41% 44%

23% 20% 40% 39% 44% 41% 38% 35% 41% 44% 39% 44% 42% 34% 39% 38% 39% 38%

8% 9% 17% 16% 18% 18% 17% 19% 17% 19% 24% 24% 21% 21% 21% 21% 20% 18%

88% 79% 84% 84% 88% 82% 80% 75% 77% 80% 77% 84% 88% 87% 83% 88% . 83%

69% 69% 74% 74% 75% 68% 60% 62% 62% 64% 66% 76% 78% 75% 73% . . 70%

51% 55% 59% 52% 48% 45% 42% 41% 43% 50% 53% 60% 61% . . . . 51%

This table presents the results from estimations of firm-level accrual mispricing over time. It shows the percentage of firms each year where significant accrual mispricing is present. Firmlevel mispricing is calculated as a rolling figure over 25–31 years. That is, we require at least 25 years of continues data for a firm to be included in this sample. For instance, to estimate mispricing for 1995, the firm needed to have financial data available for at least 25 years in the period 1965–1994 and so on. A '.' indicates years where there is not enough data-years available to estimate mispricing persistence.

This table presents the returns for a strategy of longing significantly underpriced firms and shorting significantly overpriced firms for each year in the period 1995–2011, in relation to the investigation of persistence of firm-level accrual mispricing. The returns to the long/short strategy are calculated as industry-adjusted returns for a buy and hold strategy over each year. The overall yearly abnormal returns are the difference in the long/short strategy. The bottom row of the table presents the returns to a dollar invested in the $1 long/$1 short strategy cumulatively over the period investigated. Note: Firms are considered to be significantly over or underpriced when their mispricing is significant at least at the 10% level.

−0.186 −0.226 0.039 $2.01 −0.331 −0.522 0.191 $1.93 0.019 −0.164 0.183 $1.67 −0.390 −0.469 0.079 $1.66 0.012 −0.082 0.094 $1.45 0.045 −0.057 0.102 $1.33 −0.059 −0.209 0.149 $1.23 $1 long $1 short Abnormal return (%) Invest $1 in 1995

−0.014 −0.088 0.074 $1.07

−0.003 −0.021 0.018 $1.26

0.046 0.118 −0.073 $1.16

−0.126 −0.161 0.034 $1.20

−0.019 −0.084 0.064 $1.54

−0.068 −0.010 −0.058 $1.57

−0.224 −0.189 −0.035 $1.52

−0.216 −0.149 −0.066 $1.42

−0.049 −0.018 −0.031 $1.62

2010 2009 2008 2007 2006 2005 2004 2003 2002 2001 2000 1999 1998 1997 1996 1995

Table 5 Abnormal returns from a portfolio of all significantly mispriced shares.

−0.001 0.035 −0.036 $1.93

M. Strydom et al. / International Review of Financial Analysis 34 (2014) 157–165

2011

162

Table 6 Characteristics of significantly mispriced firms. Panel A: sector breakdown of significantly mispriced firms Sector Energy Materials Industrials Consumer discretionary Consumer staples Health care Information technology Telecommunication services Utilities Total

% of sample firms that are mispriced in each sector 7.92% 8.33% 19.32% 19.32% 6.38% 12.81% 18.18% 2.34% 5.37% 100%

Percentage of firms in the market in sector 8.03% 9.31% 15.53% 20.74% 5.24% 13.18% 19.81% 2.98% 4.18% 100%

This table presents the sector breakdown of significantly mispriced firms identified for the period 1995–2011. The percentage of firms listed in each sector investigated in this study is presented first, followed by the percentage of firms in that sector in the market. Financial firms are excluded from the analyses since they are not included in the accrual mispricing sample, as discussed earlier. Panel B: Size of significantly mispriced and non-significantly mispriced firms Total assets (in millions) Year

Not significantly mispriced

Significantly mispriced firms

1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 Average

$1450.231 $1543.847 $1661.916 $1725.138 $2040.792 $2486.235 $2895.709 $2932.774 $3079.706 $3301.902 $3353.333 $3773.577 $4273.846 $4390.092 $4465.261 $4881.454 $5573.216 $3166.41

$2058.101 $2319.08 $4150.904 $3686.804 $4448.256 $6500.361 $5117.096 $6046.717 $5558.091 $7641.785 $10,824.57 $6624.927 $7931.181 $11,214.29 $12,503.4 $13,978.79 $14,528.66 $7360.77

This table presents the average size for both non-mispriced and mispriced firms for the period 1995–2011. Firms were previously identified as significantly mispriced or otherwise. Firm size is measured as the average total asset value of firms (in millions).

of all firms in the market with those of significantly mispriced firms. It reveals a substantial difference between the percentage of industrial sector firms (19.32%) in the mispriced sample compared to those in the market overall (15.53%). Industrial sector firms appear overrepresented among the significantly mispriced. The percentages of most other sectors are very similar to that of the market. Whilst most accrual mispricing studies control for size, Palmon, Ephraim, and Yezegel (2008) propose that firm size may nevertheless still play a role in the anomaly. The size of mispriced versus nonmispriced firms is therefore investigated next. As reported in Panel B of Table 6, there appears to be substantial differences in the size (as measured by the value of total assets in millions) of mispriced and non-mispriced firms. Mispriced firms are much larger than other firms in every sample year. The average size of mispriced firms is more than twice that of non-mispriced firms over the sample period and size therefore appears to play a role in firm-level mispricing with large firms being more likely to be mispriced. Finally, the country-level accrual anomaly literature proposes a role for analysts in the existence (and persistence) of accrual mispricing. Whilst a couple some accrual mispricing studies argue that even analysts overprice accruals (Atwood & Xie, 2010; Bradshaw, Richardson, & Sloan, 2001), others show that analysts reduce information asymmetry and pricing inefficiencies through their superior analysis skills and

M. Strydom et al. / International Review of Financial Analysis 34 (2014) 157–165

163

Table 7 Analyst recommendations for sample firms. Year

Analyst recommendations made for significantly mispriced firms

Analyst recommendations made for non-mispriced firms

Total analyst recommendations

% Change in analyst following for mispriced firms

1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011

1527 (14%) 1296 (12%) 1168 (11%) 1423 (12%) 1346 (11%) 1248 (11%) 1223 (10%) 2113 (11%) 1593 (11%) 1278 (9%) 1098 (9%) 1226 (9%) 1226 (9%) 1461 (10%) 1288 (9%) 1238 (9%) 1267 (9%)

9566 (86%) 9520 (88%) 9877 (89%) 10,616 (88%) 11,201 (89%) 10,499 (89%) 10,941 (90%) 16,484 (89%) 12,957 (89%) 12,421 (91%) 11,247 (91%) 12,420 (91%) 12,592 (91%) 12,959 (90%) 12,394 (91%) 12,048 (91%) 12,905 (91%)

11,093 10,816 11,045 12,039 12,547 11,747 12,164 18,597 14,550 13,699 12,345 13,646 13,818 14,420 13,682 13,286 14,172

. −15% −10% 22% −5% −7% −2% 73% −25% −20% −14% 12% 0% 19% −12% −4% 2%

This table presents analyst recommendation data for sample firms over the period 1995–2011. It shows the annual number (and %) of recommendations for both non-mispriced and mispriced firms. The change column indicates the year-by-year change in analyst followings for mispriced firms. If a sample firm is not followed by analysts it is assigned a value of zero for number of recommendations. A '.' indicates that data is not available to estimate a value in the particular year.

information access (Barone & Magilke, 2009; Elsharkawy & Garrod, 1996; Walther, 1997). We therefore examine the differences in analyst following for mispriced and non-mispriced firms. Specifically, we determine whether changes in analysts following are related to the existence and persistence of mispriced firms. Table 7 presents the analyst following data for sample firms (split into mispriced and otherwise). The focus is mainly on the change in analyst following. Substantial decreases in analyst following are visible from Table 7 in 1996 and 1997. When comparing this to Panel B of Table 4, an increase in mispricing persistence is visible (both for two and four-year mispricing persistence) following the decrease in analyst following. A large increase in analyst following (up 73%) occurs in 2002 and a decrease in the persistence of accrual mispricing is again visible following this period in Panel B of Table 4. It therefore seems as if analyst following decreases the pervasiveness of firm-level accrual mispricing (perhaps as they discover the mispricing earlier due to better skills and information access). This finding is consistent with the suggestions of Elsharkawy and Garrod (1996), Walther (1997) and Barone and Magilke (2009). The extant literature proposes a role for earnings quality in the accrual anomaly (Drake et al., 2009). More specifically, it suggests that

mispriced firms are those with low earnings quality, perhaps due to earnings management given that investors misprice the discretionary component of accruals (Xie, 2001). Others show that earnings quality improved post-SOX, which should have reduced accrual mispricing (Cohen, Dey, & Lys, 2008; Green et al., 2011; Lobo & Zhou, 2010). However, Cohen et al. (2008) show that managers have shifted away from accrual-based earnings management in the post-SOX period given the extensive penalties imposed. They conclude that this change would likely provide management incentive to manage earnings through “real” activities such as sales promotions and decreases in discretionary expenses. We therefore next determine whether significantly mispriced firms have higher accruals, and whether they are more likely to employ real activity based earnings management post-SOX. Following the methodology in Zang (2012) and Cohen and Zarowin (2010) we compute two comprehensive measures of real earnings management which considers abnormal levels of production, discretionary expenses and cash flows. The methodology employed to estimate these are the same as in Cohen and Zarowin (2010). We first compare the level of accruals for significantly vs. non-significantly mispriced firms, but find no difference between the pre and post-SOX periods. Results are

Table 8 Comparison of total accruals and real earnings management proxies' pre and post-SOX for significantly mispriced firms. Pre-SOX

Post-SOX

Mean

Std Dev

Mean

Std Dev

Panel A: total accruals Significantly mispriced vs. Non-mispriced firms t-Test difference Significantly Overpriced vs. Underpriced firms t-Test difference

−0.040 −0.046 (t = −0.41, p = .68) −0.043 −0.047 (t = 3.18, p = .001)

0.057 0.059

−0.036 −0.037 (t = 0.34, p = .74) −0.036 −0.038 (t = 1.03, p = .30)

0.053 0.056

Panel B: real earnings management Significantly mispriced vs. Non-mispriced firms t-Test difference Significantly overpriced vs. Underpriced firms t-Test difference

−0.086 −0.081 (t = −1.10, p = .27) −0.093 −0.076 (t = −2.33, p = .02)

0.125 0.117

−0.076 −0.083 (t = 2.05.10, p = .04) −0.081 −0.072 (t = −1.17, p = .24)

0.11 0.117

0.057 0.056

0.137 0.109

0.053 0.054

0.122 0.102

This table presents the results from tests comparing the level of accruals and real earnings management for significantly over and underpriced firms in the pre and post-SOX periods as identified in Table 4 earlier. Panel A presents the results for total accruals whilst Panel B documents that for real earnings management. Significant results from t-tests are in bold face. Real earnings management was estimated with two comprehensive measures (abnormal production and abnormal discretionary expenses) as in Roychowdhury (2006) and data for this was obtained from the Compustat database.

164

M. Strydom et al. / International Review of Financial Analysis 34 (2014) 157–165

presented in Table 8.10 Next, we investigate those firms earlier identified as significantly over or underpriced to determine whether any difference exists in the level of their total accruals. A t-test confirms that significantly overpriced firms have higher levels of accruals in the pre-SOX period when compared to underpriced firms (t-stat = 3.18, p = .001). Post-SOX, however, there is no longer a significant difference between the accrual levels of these significantly over and underpriced firms, indicating perhaps that accruals (and particularly the discretionary component of accruals) decreased postSOX. Turning to the likelihood of real earnings management (REM) in these firms, we first show that the likelihood of underpriced firms involving in REM is substantially higher in the pre-SOX period compared to overpriced firms (t-stat = −2.33, p = .02). In the post-SOX period, however, there are no significant differences in the likelihood of over and underpriced firms involving in REM. As the option of accrual-based earnings management is limited post-SOX this result likely indicates that all firms who manage earnings do so through real activities in this period. This conjecture is supported further by tests comparing the likelihood of all significantly mispriced firms (compared to those that are accurately priced) involved in REM post-SOX (see t-test results in Panel B of Table 8). Results show that mispriced firms are significantly more likely to manage earnings (through real activities management) post-SOX than accurately priced firms (t-stat = 2.05, p = .04). This indicates, consistent with Cohen et al. (2008), that whilst earnings quality may have improved post-SOX, managers are still managing earnings — only in a different way. Our results suggest that investors are equally unable to distinguish the pricing implications of such REM given that mispricing (both over and underpricing) of accruals continues in this period.

the firm-level, especially when attempting to determine its cause. Finally, we document that firm-level accrual mispricing persists long enough for investors to potentially benefit from trading on it. A firm-level accruals-based trading strategy, buying underpriced accrual firms and selling overpriced ones, yields abnormal returns consistent with country-level accrual anomaly findings. Future studies could estimate the portfolio profitability of trading on firm-level accrual mispricing. Our findings have implications for investors, firms, and regulators. Whilst certain firms are overpriced, as the accrual anomaly predicts, others are underpriced. Investors should therefore identify these specific firm positions in their pricing decisions. In addition, as mispriced firms may remain mispriced for more than one period, investors who can identify such firms could (in theory) profit from a strategy of buying underpriced firms and selling overpriced ones. For firms it is important to note that investors misprice individual securities. Improved disclosure quality should reduce that information asymmetry and result in more accurate pricing. Finally, whilst increased disclosure regulation at the country-level reduces such mispricing, regulators should note that firm-level mispricing remains. Further regulatory reforms on improving information disclosures may be needed to reduce such mispricing.

5. Conclusion and discussions

Ali, A., & Gurun, U. G. (2009). Investor sentiment, accruals anomaly and accruals management. Journal of Accounting, Auditing and Finance, 24, 415–431. Atwood, T. J., & Xie, H. (2010). The market mispricing of special items and accruals: One anomaly or two? Review of Accounting Studies, 9, 156–179. Barone, G. J., & Magilke, M. J. (2009). An examination of the effects of investor sophistication on the pricing of accruals and cash flows. Journal of Accounting, Auditing and Finance, 24, 385–414. Beneish, M., & Vargus, M. (2002). Insider trading, earnings quality and accrual anomaly. Accounting Review, 77, 755–791. Bhojraj, S., & Swaminathan, B. (2009). How does the corporate bond market value capital investments and accruals? Review of Accounting Studies, 14, 31–62. Bradshaw, M. T., Richardson, S. A., & Sloan, R. G. (2001). Do analysts and auditors use information in accruals? Journal of Accounting Research, 39, 45–74. Cohen, D., Dey, A., & Lys, T. (2008). Real and accrual based earnings management in the pre and post Sarbanes Oxley periods. Accounting Review, 83, 757–787. Cohen, D., & Zarowin, P. (2010). Accrual-based and real earnings management activities around seasoned equity offerings. Journal of Accounting and Economics, 50, 2–19. Collins, D. W., & Hribar, P. (2000). Earnings-based and accrual-based market anomalies: One effect or two? Journal of Accounting and Economics, 29, 101–123. De Fond, M., & Park, C. (2001). The reversal of abnormal accruals and the market valuation of earnings surprises. The Accounting Review, 76, 375–404. Desai, H., Rajgopal, S., & Venkatachalam, M. (2004). Value-glamour and accruals mispricing: One anomaly or two? Accounting Review, 79, 355–385. Dopuch, N., Seethamraju, C., & Xu, W. H. (2010). The pricing of accruals for profit and loss firms'. Review of Quantitative Finance and Accounting, 34, 505–516. Drake, M. S., Myers, J. N., & Myers, L. A. (2009). Disclosure quality and the mispricing of accruals and cash flow. Journal of Accounting, Auditing and Finance, 24, 284–357. Elsharkawy, A., & Garrod, N. (1996). The impact of investor sophistication on price responses to earnings news. Journal of Business Finance and Accounting, 23, 221–236. Fairfield, P.M., Whisenant, J. S., & Yohn, T. L. (2003). Accounting Review, 78, 353–371. Fama, E. F. (1998). Market efficiency, long-term returns, and behavioral finance. Journal of Financial Economics, 49, 283–306. Green, J., Hand, R. M., & Soliman, M. (2011). Going, going, gone? The apparent demise of the accruals anomaly. Management Science, 57, 797–816. Hirshleifer, D., Teoh, S. H., & Yu, J. J. (2011). Short arbitrage, return asymmetry, and the accrual anomaly. Review of Financial Studies, 24, 2429–2461. Iliev, P. (2010). The effect of SOX section 404: Costs, earnings quality, and stock prices. Journal of Finance, 62, 1163–1196. Keskek, S., Myers, L. A., Omer, T. C., Sharp, N. Y., Keskek, S., Myers, L. A., et al. (2013). Exploring the accrual-related optimism in management earnings forecasts. Working paper (Available at SSRN: http://ssrn.com/abstract=1975122). Kraft, A., Leone, A. J., & Wasley, C. (2006). An analysis of the theories and explanations offered for the mispricing of accruals and accrual components. Journal of Accounting Research, 44, 297–339. Kraft, A., Leone, A. J., & Wasley, C. (2007). Regression-based tests of the market pricing of accounting numbers: The Mishkin test and ordinary least squares. Journal of Accounting Research, 45, 1081–1114.

This study investigated the mispricing of accruals at the firm-level. It specifically examined whether firm-level accrual mispricing differs from the country-level mispricing documented by Sloan (1996). It then considered the behavior of mispriced firms over time and, more specifically, whether firm-level mispricing persists. It also determined whether abnormal returns are available from a trading strategy based on buying underpriced firms and selling overpriced firms. Finally it examined the industry, firm size, analyst following and extent of real earnings management of significantly mispriced firms to determine whether any of these factors were related to mispricing. The firm-level investigation of firm-level accrual mispricing revealed both significant under and overpricing of accruals. This mispricing remained persistent with most firms identified as mispriced in oneyear remaining so in the next. Furthermore, at least half of mispriced firms remained so for four or more years. An accrual-based trading strategy buying significantly underpriced firms and selling overpriced ones yielded substantial abnormal returns, indicating that investors could profit from this mispricing. Lastly significantly mispriced accrual firms were more common in the industrial sector and tended to be larger firms. Analyst following seems important in limiting accrual mispricing persistence with greater analyst following being inversely associated with mispricing. It also appears that post-SOX mispriced firms are managing earnings mainly through real earnings management activities. This study contributes to accrual mispricing literature by first documenting that firm-level accrual mispricing differs from the well documented country-level anomaly. It is the first study to do so. It therefore establishes the importance of investigating the accrual anomaly at

10 Results obtained using the two comprehensive real earnings management proxies from Roychowdhury (2006) are very similar and only the results from the first proxy (including abnormal discretionary expenses and abnormal production costs) are therefore reported in Table 8.

Acknowledgments The authors would like to thank participants at the 2012 AAA meetings in Washington and the 2013 SFA meetings in San Juan for their useful comments on this paper. All errors remain our own. References

M. Strydom et al. / International Review of Financial Analysis 34 (2014) 157–165 Lev, B., & Nissim, D. (2006). The persistence of the accrual anomaly. Contemporary Accounting Research, 23, 193–226. Lobo, G. J., & Zhou, J. (2010). Changes in discretionary financial reporting behavior following the Sarbanes–Oxley Act. Journal of Accounting, Auditing and Finance, 25, 1–26. Mashruwala, C., Rajgopal, S., & Shevlin, T. (2006). Why is the accrual anomaly not arbitraged away? The role of idiosyncratic risk and transaction costs. Journal of Accounting and Economics, 42, 3–33. Mishkin, F. (1983). A rational expectations approach to macroeconometrics: Testing policy effectiveness and efficient-markets models. Chicago, IL: University of Chicago Press. Palmon, D., Ephraim, F. S., & Yezegel, A. (2008). The accruals anomaly and company size. Financial Analysts Journal, 64, 47–60. Rangan, S., & Sloan, R. G. (1998). Implications of the integral approach to quarterly reporting for the post-earnings announcement drift. Accounting Review, 73, 353–371. Richardson, S., Tuna, I., & Wysocki, P. (2010). Accounting anomalies and fundamental analysis: A review of recent research advances. Journal of Accounting and Economics, 50, 410–454.

165

Roychowdhury, S. (2006). Earnings management through real activities manipulation. Journal of Accounting and Economics, 42, 335–370. Sarbanes–Oxley Act of 2002, Pub. L. No. 107–204, 116 Stat. 745. Retrieved from http:// www.sec.gov/about/laws.shtml#sox2002. (2002). Shi, L., & Zhang, H. (2012). Can the earnings fixation hypothesis explain the accrual anomaly? Review of Accounting Studies, 17, 1–21. Singer, Z., & You, H. (2011). The effect of section 404 of the Sarbanes–Oxley Act on earnings quality. Journal of Accounting, Auditing and Finance, 26, 556–589. Sloan, R. G. (1996). Do stock prices fully reflect information in accruals and cash flows about future earnings? Accounting Review, 71, 289–315. Walther, B. R. (1997). Investor sophistication and market earnings expectations. Journal of Accounting Research, 35, 157–179. Xie, H. (2001). The mispricing of abnormal accruals. Accounting Review, 76, 356–373. Zang, A. (2012). Evidence on the trade-off between real activities manipulation and accrual-based earnings management. Accounting Review, 87, 675–703.