Intangible Returns, Accruals, and Return Reversal: A Multiperiod Examination of the Accrual Anomaly

Transcription

1 THE ACCOUNTING REVIEW Vol. 85, No pp Intangible Returns, Accruals, and Return Reversal: A Multiperiod Examination of the Accrual Anomaly Robert J. Resutek Dartmouth College DOI: /accr ABSTRACT: Prior studies employ a two-period empirical model and interpret the negative association between accruals in period one and returns in period two as evidence that investors misprice the information contained in accruals. In contrast to prior studies, I employ a three-period log-linear model decomposed from a firm s book-to-market ratio and show that investors do not misprice the information contained in accruals. My study shows that in the four-year period prior to accrual recognition, equity prices tend to be driven disproportionately by intangible returns, or returns not explained by accounting measures. Accordingly, the relation between prior period intangible returns and future period returns subsumes the relation between current period accruals and future returns. In addition, I link the accrual anomaly and the value/growth anomaly to a common economic mechanism intangible returns and show that a strong negative relation between external financing activities and future returns is not subsumed by the accrual anomaly. Keywords: accrual anomaly; book-to-market; external financing; investment; market efficiency. Data Availability: The data used are available from the sources noted in the study. I. INTRODUCTION Beginning with Sloan 1996, the accounting literature has produced an extensive stream of literature examining the negative relation between current period accruals and future period returns. From this stream of literature two primary explanations for the negative relation the accrual anomaly have emerged. The persistence explanation builds off Sloan 1996 and explains the accrual anomaly as caused by investor inability to distinguish the pricing This paper is based on my dissertation at the University of Texas at Austin. I would like to thank the members of my committee: Robert Freeman chair, Michael Clement, Ross Jennings, Sheridan Titman, and Yong Yu for their helpful comments and suggestions on earlier drafts of this paper. In addition I thank Rich Frankel, Kenneth French, Bill Kinney, Jonathan Lewellen, William Mayew, Richard Sansing, two anonymous reviewers, Mark Trombley editor, and seminar participants at Washington University, Dartmouth, Stanford, Michigan State University and participants of the McCombs School of Business summer brown bag series. I am grateful for the financial support of the Deloitte & Touche Foundation, McCombs School of Business, and Tuck School of Business at Dartmouth. Editor s note: Accepted by Mark Trombley. Submitted: July 2008 Accepted: January 2010 Published Online: June

2 1348 Resutek implications of the differential persistence of the accrual and cash flow components of current earnings. The growth explanation builds off Fairfield et al and suggests that investors misprice the growth information contained in accruals. While these two streams of literature are distinct, they share some similarities. First, both explanations suggest that investors may be subject to behavioral biases that prevent them from properly incorporating publicly available accounting information into stock prices. 1,2 Second, as noted by Fairfield et al and Zhang 2007, large positive accruals tend to occur during periods of growth and equity issuance. There is a strong negative relation between equity issuance activity and future returns Loughran and Ritter Despite the empirical relation between equity issuance, accruals, and future returns, accrual studies in both streams have been relatively silent as to the effect that the interaction of these three variables on inferences made between future period returns and current period accruals. 3 Finally, both streams examine the relation between accruals and future returns primarily within the confines of a strict two-period model. 4 For example, studies in these streams generally examine whether variation in current period accruals are associated with variation in future returns after controlling for current period characteristics and factors known to be associated with future returns. This study takes a different approach from that taken in prior accrual studies and finds results suggestive of a third explanation. Prior research has viewed a firm s book-to-market ratio B/M as a characteristic distinct from the firm s current period accruals. However, this view overlooks the fact that current period accruals affect B/M. Consider that B/M is comprised of four primary elements: 1 retained earnings, 2 invested equity, 3 expected cash flows, and 4 expected returns. Accordingly, changes in each of these elements explain changes in B/M. Given that accruals explain part of the changes in retained earnings, and that changes in each of the other three elements are associated with future returns, this study examines whether the evolution of B/M can shed new light on the negative association between current period accruals and future period returns. The empirical design employed in this study builds on the work of Daniel and Titman 2006; hereafter, DT. DT note that the information investors incorporate into price can be partitioned into two parts: tangible information and intangible information. By assuming that a firm s market return over a given period is a good proxy for total information, DT classify the portion of a firm s realized market return that is explained by summary accounting measures as tangible information, or tangible returns. Accordingly, the residual realized return variation that is not explained by summary accounting measures is termed intangible information, or intangible returns. DT show in a two-period empirical model that the strong negative relation between price-scaled accounting variables e.g., book-to-market, earnings-to-price and future returns is not associated with total realized growth as measured by accounting variables. Rather, DT suggest that the return reversal associated with price-scaled accounting variables is driven by realized intangible returns. I expand DT s empirical model into a three-period model to investigate the relation between prior period intangible returns in period one, accruals in period two, and future returns in period Assuming certain conditions relating to changes in a firm s expected returns and optimal investment in response to these changes, the growth explanation does not necessarily imply investor irrationality. Recent work by Mashruwala et al and Khan 2008 suggest rational asset-pricing explanations of the accrual anomaly. Specifically, Mashruwala et al. suggest that arbitrageurs are prevented from capitalizing on the mispricing associated with accruals due to high arbitrage risk, while Khan proposes a four-factor ICAPM-based pricing model that explains the cross-sectional future return variation attributable to the current period accruals. Recent work by Cohen and Lys 2006 and Dechow et al suggest that the negative relation between future returns and accruals subsumes the negative relation between future returns and external financing activities. Kothari et al and Zhang 2007 consider the relation between accruals in the current year and firm growth in prior years, the current year, and subsequent years.

3 Intangible Returns, Accruals, and Return Reversal 1349 three. Intangible returns can generally be interpreted as capturing future growth options that have not been recognized by accounting performance measures. Given the discrete, conservative nature of financial accounting, intangible returns will tend to lead tangible returns as the market prices information more quickly than GAAP recognizes value. By examining the relation between current period accruals and prior period intangible returns, derived from the book-to-market ratio, I provide insight into a possible common link between the value/growth anomaly and the accrual anomaly. Consistent with prior studies, I show that in two-period settings, the accrual component of log book returns explains a significant portion of future return variation. Controlling for firm growth captured by accounting measures of growth over the four-year period prior to accrual recognition, I show that the accrual component of log book returns still explains significant future return variation. However, once I control for prior period intangible returns PPIR, I show that the return reversal previously attributed to accruals is subsumed by PPIR. More germane to the accounting literature, I show that once intangible returns from period one are controlled, the negative relation between accrual-based returns in period two and future returns in period three is insignificant. This pattern holds for multiple definitions and decompositions of accruals. This study makes several contributions to the existing literature examining the relation between future returns and accounting measures of performance such as accruals. First, I provide empirical evidence that investors correctly incorporate the information contained in accruals into price. This evidence is significant given the widely held view that investors misprice the growth information contained in accruals Fairfield et al. 2003; Zhang 2007 and/or the persistence characteristics of accruals Sloan 1996; Richardson et al Second, this study contributes to research examining the relation between external financing activities, accruals, and future returns. Recent studies examining the relation between these three variables are mixed. Cohen and Lys 2006 and Dechow et al suggest that the negative relation between future returns and external financing activities is essentially subsumed by the accrual anomaly, while Daniel and Titman 2006 show that their composite issuance variable subsumes any relation between future returns and accounting-based performance measures. The empirical evidence reported in this study serves to link these two disparate findings. Specifically, this study highlights an economically significant negative relation between future returns and equity issuance/repurchase activity and intangible returns while finding no significant association between future returns and accruals. Third, this study employs an empirical design that addresses some of the concerns raised by Kraft et al. 2006, 2007 relating to accrual anomaly studies. Specifically, Kraft et al show how the exclusion of variables correlated with accruals can affect inferences with future returns from Mishkin 1983 pricing tests. The empirical design used in this study endogenizes the pricing of all prior and current period information. This empirical design structure is important for two reasons. First, it mitigates the concern that significant relations between explanatory variables and future returns are caused by the omission of relevant variables. Second, the inferences drawn between future period returns and current period accruals are not contingent on the rational pricing of all information. Despite the agnostic stance on rational pricing this empirical design allows, the design still allows for inferences to be drawn about whether accruals are properly incorporated into price. Fourth, this study provides a link between the value/growth anomaly and the accrual anomaly. Although the theories used to explain both anomalies are similar, these anomalies have generally been considered distinct. This study extends the idea of Beaver 2002 and Desai et al who suggest that the accrual anomaly may be a function of the value/growth anomaly, providing the economic intuition and empirical evidence that shows these two anomalies are likely driven by the same underlying construct specifically, intangible returns.

4 1350 Resutek Section II introduces the empirical design and the economic intuition supporting it. Section III provides empirical evidence examining the relation between PPIR, current period accruals, and future returns. Sections IV and V discuss empirical evidence from additional robustness tests. Section VI provides concluding remarks. II. RESEARCH DESIGN METHODS AND INTUITION Research Motivation and Variable Intuition The value/growth anomaly is characterized by the positive association between future returns and price-scaled accounting variables such as book-to-market and earning-to-price. One of the more prominent explanations for the value/growth anomaly, proposed by Lakonishok et al. 1994; hereafter, LSV, suggests that the future return patterns of value and growth firms are due to investors over-extrapolating realized accounting measures of growth. LSV suggest that investors over-react to realized earnings and sales, bidding the price of firms with high past accounting growth too high. In subsequent periods when investor expectations of sales and earnings are not met, the book-to-market ratios of growth firms increase as investors reassess their earnings expectations and bid prices down. In a recent study, Daniel and Titman 2006 provide empirical evidence countering the claims of LSV. DT show that past accounting growth, when measured on a per-share basis, is unrelated to future return patterns. Further, DT show that the future return pattern LSV associate with five-year past accounting growth is actually due to information that is unrelated i.e., orthogonal to past accounting returns. This information, tagged intangible returns by DT, is strongly associated with future returns over a five-year period and can be viewed as a proxy for changes in investor expectations of future earnings and/or changes in firm risk. The empirical evidence of DT raises the question as to whether inferences between other accounting measures associated with firm growth and future returns are also affected by intangible returns and/or equity issuance activity. Of particular interest to the accounting researchers over the past 15 years is the negative relation between future period returns and current period accruals, commonly known as the accrual anomaly. The primary intuition used to explain the accrual anomaly is similar in many respects to that proposed by LSV. Specifically, firms with high low current accruals tend to experience lower higher future earnings and, because investors do not understand this relation, lower higher returns are realized in future periods. However, while the intuition used to explain the future return pattern of the value/growth anomaly is similar to that used to explain the accrual anomaly, the empirical patterns are distinct. Whereas the value/growth anomaly is relatively long in horizon up to five years, the accrual anomaly is relatively short, with current period accruals only explaining future return variation for about one year. In addition, prior accrual studies show that simple controls for value/growth characteristics B/M or risk factors do not explain the accrual anomaly. Accordingly, with the exception of Desai et al. 2004, most studies have viewed the value/growth anomaly and the accrual anomaly as distinct. Examining the evolution of B/M offers an alternative and possible improvement to the twoperiod research designs used in prior accrual studies, while examining possible common links between the accrual anomaly and value/growth anomaly. By construction, B/M represents the intersection of four variables: 1 retained earnings, 2 equity investment, 3 expected future earnings, and 4 expected returns. Accordingly, changes in B/M capture changes in at least one of these elements. While the accounting literature has focused on understanding the relation between accrual-based changes in retained earnings and future returns, the finance literature has focused on understanding the relations between changes in the other three elements of B/M and future returns. For example, equity issuance and repurchase activity affect the equity base of the firm, and prior studies have shown that these activities are associated with future returns for up to five years Loughran and Ritter 1995; Ikenberry et al Further, the primary argument in the value/

5 Intangible Returns, Accruals, and Return Reversal 1351 growth literature is whether the positive association between future returns and price-scaled accounting variables is due to reversals of irrational future earnings expectations formed in prior periods LSV; Barberis et al or rational shifts in risk Fama and French 1992, Prior accrual studies have not necessarily ignored the effects that changes in the other elements of B/M have on inferences between future returns and accruals. Most studies use B/M as a control variable for characteristics associated with firm growth and expected returns. However, it is possible that B/M is too noisy of a variable to control for effects on future returns due to changes in equity issuance, changes in expected earnings, and shifts in risk Fama and French To illustrate, consider a firm that experiences a positive shock to expected cash flows in period one, the period prior to the recognition of abnormally large accruals. Due to the discrete, conservative nature of accounting, the shock to future cash flows will tend to be reflected in market value earlier than book value, especially for positive shocks, and B/M will be lower at the end of period one. 5 In the subsequent period, period two, managers capitalize on these positive NPV projects and expand productive capacity. To the extent that managers do not fund these projects with existing cash or debt facilities, managers may issue equity. With this decision to expand productive capacity, positive accruals are recognized. Since accruals are just one element of the four comprising B/M, properly controlling for changes in the other three elements is necessary to distill the incremental future return variation that is explained by accruals from that explained by changes in the other B/M elements occurring in periods contemporaneous with, and prior to, the recognition of accruals. A decomposition of current B/M into different components offers a structural way to define distinct variables that capture changes in each of these elements. In log form, the book-to-market ratio at time t is equal to the book-to-market ratio at time t plus log change in book value, minus log change in market value. log B t /M t = bm t = bm t + log B t /B t log M t /M t = bm t + log B t,t log M t,t. 1 Log change variables have convenient additive properties that allow a cumulative multi-period change variable to be decomposed into multiple single-period change variables. Using this property, bm t can be decomposed into the following five year decomposition: bm t = bm t 5 + log B t 1 B t 5 + log B t B t 1 log M t M t 5 ; 2 Bt bm t = bm t-5 + log Bt 1 5 NOAt + log NOA t 1 Bt + log NOAt M log B t 1 M NOAt 1 t t 5 accruals TACC non accrual growth N_TACC total log change inbt 3 5 This tendency is documented by Kothari et al. 2006, Table 2. Their evidence shows that firms classified as high-accrual firms tend to experience increasingly larger abnormal returns in the years leading up to classification as a high-accrual firm. Complementing this evidence, they show that firms classified as low-accrual firms tend to experience increasingly negative abnormal returns in the years leading up to classification as a low-accrual firm, although the absolute magnitude of the returns is smaller relative to the high-accrual firms.

6 1352 Resutek The above decomposition 3 shows that bm t is equal to bm t 5 plus several sub-period, accountingbased log changes in value, minus a cumulative market-based log change. Note that logarithmic properties allow me to decompose log change in book value between t 1 and t into the accrual and non-accrual component of log book growth. The definition of TACC is consistent to accruals as described in Feltham and Ohlson 1995 and with total accrual variables used in Richardson et al. 2005, 2006, Dechow et al. 2008, and other recent studies examining total accruals. 6,7 N_TACC is mathematically equivalent to log B t 1,t log NOA t 1,t. Assuming the firm has no equity issuances/repurchases between t 1 and t, change in book value equals net income, and N_TACC can be interpreted in a manner similar to free cash flow, the cash component of earnings that Dechow et al suggest investors also misprice. For robustness purposes, alternative accrual and non-accrual definitions are examined in later sections. Log Returns versus Log Changes The log-linear decomposition of the book-to-market ratio is useful because it gives simple variables that can proxy for the dynamics of B/M evolution discussed above. Consistent with DT, I interpret changes in accounting values as representing tangible information e.g., TACC, N_TACC. To proxy for intangible information, or information not captured by summary accounting measures, I need to construct a variable that represents information used by investors that is unrelated to accounting information. To do this, I regress realized log market returns over a given period of time on tangible return components of B/M over that same time period. Assuming log market return is a good proxy for total value-relevant information, I interpret the fitted value from this regression as tangible information and the residual as intangible information. Before running these regressions, log change values defined in Equation 3 above need to be transformed into log return values. Assuming there are no equity issues, dividends, or other corporate actions affecting shareholder s equity, log change in market value between t and t is equal to the log return an investor earns between t and t i.e., log M t,t = ret m t,t. However, as the time between t and t increases, log M t,t will generally not equal ret m t,t. To convert log change in market value into a log return metric, I introduce the variable iss t,t. This composite equity issuance variable, iss t,t, is defined consistent with the composite equity issuance variable t,t of DT and represents the difference between log market return and total log change in market value. More specifically: iss t,t = log M t,t ret m t,t. 4 The iss t,t variable can be used to convert log B t,t into a log return measure. For example, just as ret m t,t = log M t,t iss t,t, a log book return variable can be constructed as ret b t,t = log B t,t iss t,t. The variable ret b t,t gives the log book value at t of a $1 investment in book equity made at t. Since ret b t,t is equal to log B t,t minus iss t,t, the one-year log book return between t 1 and t can be defined as follows: ret b t 1,t = log B t 1,t iss t 1,t = TACC + N_TACC iss t 1,t Richardson et al. 2005, 2006 argue that change in net operating assets offers a more complete measure of accruals. For robustness, this study also considers the more common working capital accruals in Table 4. Prior studies have generally defined accruals as the linear change in accrual accounts, scaled by total assets. This study defines accruals as the log change in the accrual account e.g., log NOA t 1,t, log Working Capital t 1,t. Conceptually, both types of accrual definitions capture the dynamics associated with one-year growth in net operating assets or net working capital. However, the logarithmic definition can be interpreted as the log return earned by equity holders i.e., the accrual component of earnings once equity issuance is controlled. For robustness purposes, the linear variable is directly examined in Table 5.

7 Intangible Returns, Accruals, and Return Reversal 1353 As Equation 5 reflects, the log book return between t 1 and t can be interpreted as the sum of three log return components: total accruals TACC, non-accrual growth N_TACC, and equity issuance. 8 When these three components are used as explanatory variables of future monthly returns, the coefficient on TACC will represent the expected average monthly return in period t+1 for an investor who realized $1 of log accrual return between period t 1 and t, holding constant log book return attributable to N_TACC and iss t 1,t. Alternatively, the coefficient on the iss t 1,t variable will capture the expected average monthly return in period t+1 for an investor who invested $1 in the new equity issued at t 1. 9 Intangible Value Change Once log change in book value log B t,t is transformed into a log book return ret b t,t, I construct a proxy for intangible information by employing a simple process of estimating crosssectional annual regressions, defining the residuals from each regression as intangible returns. More specifically, to compute the change in value due to intangible information over the one year current period t 1, t and the four year prior period t 5, t 1, I regress the realized log market returns inclusive of dividends, ret m i, t,t,onbm i, t and the components of the ret b i, t,t over the same period of time t, t. ret m i, t 1,t = bm i, t TACC i + 3 N_TACC i + 4 iss i, t 1,t + u i,t ; ret m i, t 5,t 1 = bm i, t log B i, t 5,t iss i, t 5,t 1 + u i,t 1. 6a 6b Consistent with DT, I define the return due to tangible information as the fitted component and intangible information as the residual from the above regressions: 10 u i,t = CPIR current period intangible return; and u i,t 1 = PPIR prior period intangible return. By construction, total log return is equal to the sum of tangible returns plus intangible returns. 11 III. RESULTS Data Definitions and Construction My study examines the relation between future returns and accruals in firms with fiscal year-ends between 1/31/67 and 12/31/06. To be included in the sample, firms must appear on the CRSP/Compustat merged database and have positive book value of equity at fiscal year-end for 8 9 There are two alternative methods for transforming total log change in book value into ret b t,t. First, retb t,t can be computed using a share adjustment factor as noted in the CRSP/Compustat data manual. Second, iss t,t can be computed as a plug: iss t,t = bm t bm t log B t + ret m t. An important subtlety to note is that TACC and N_TACC can be interpreted as accrual and non-accrual components of earnings in multivariate settings that include iss t 1,t. Variation in log B t 1,t is explained by earnings for that period once equity issuance and repurchase activity is controlled. Hence, in multivariate settings that include iss t 1,t, TACC can be viewed as the total accrual component of earnings. 10 Fitted values for 6a and 6b explain 17 percent and 40 percent of the cross-sectional variation of ret m t 1,t and retm t 5,t 1, respectively. 11 Another interpretation of iss t,t is that this variable captures elements of intangible information not captured by PPIR or CPIR. By construction, the intangible returns defined above CPIR and PPIR only capture the intangible information that investors use to price securities. Accordingly, it is reasonable to assume that a manager s intangible information is more precise and extensive than the empirical proxies defined above. Prior literature Hovakimian et al suggest that firms are more likely to issue repurchase equity following periods when stock prices increase decrease more dramatically than earnings. Accordingly, iss t,t can be interpreted as an additional and possibly independent measure of intangible information.

8 1354 Resutek years t 5, t 1, and t. However, firms with negative book values in intermediate years are not excluded. In addition, firms must have positive net operating asset balances at fiscal year-end of years t 1, and t. Additional accounting measures used later in robustness tests affect the sample sizes used in those robustness tests accordingly, although not significantly. I exclude financial firms SIC , utilities SIC , firms with stock prices on CRSP of less than $5 per share as of the last day of trading in June of year t+1, and firms without a complete series of 60 monthly returns between fiscal year-end t 5 and fiscal year-end t. The future return series begins on July 1, t+1 and delisting returns are used, when necessary, to compute annual portfolio returns. 12,13 I also require that the firm have a valid price and shares outstanding as of the last day of trading in fiscal year t 5, t 1, t and the last day of trading in June of year t+1. From this final set, I annually trim 0.5, 99.5 the tangible return components TACC, WCACC, N_TACC, and N_WCACC variables. Consistent with prior studies, scaled linear change variables used in later robustness tests are annually winsorized 0.5, Returns are from CRSP. All other variables are from the CRSP/Compustat merged database. A firm s log book-to-market ratio at time t bm t is computed as the log book value of the firm at fiscal year-end of year t minus the log market value of the firm as of the last day of trading in fiscal year t as reported in CRSP/Compustat merged database data25 data199. All log change variables are defined with two subscripts: the first subscript notes the beginning of the change period and the second subscript notes the end of the change period. For example, log B t,t is computed as log book value as of the end of fiscal year t minus the log book value as of the end of fiscal year t. Similarly, log return measures e.g., ret b t,t are italicized and defined with two subscripts, with the first subscript noting the beginning of the return period and the second subscript noting the end. Data Summary Table 1 reports summary statistics. These statistics show that accruals appear to be a relatively symmetric variable, although there is some right skewness. The skewness is consistent with the average firm growing its balance sheet over time. Given the data requirements of a trailing five-year return series, my sample is skewed toward larger, more mature firms. However, the standard deviation of the variables suggests considerable variation, despite their relative homogeneous size characteristic. Table 2 reports pairwise correlation coefficients. Overall, the univariate correlations support my hypothesis that PPIR affects inferences of the relation between future returns and current period accruals. First, consistent with the prior accrual literature, each accrual measure is negatively correlated with bm t. Second, TACC is more strongly correlated with PPIR than working capital accruals WCACC. Further the relation between accrual variables and PPIR is stronger than the relation between the respective accrual variables and bm t, suggesting that bm t may be an inadequate control for prior period intangible returns. Finally, future returns Ret are negatively related to each accrual measure as well as PPIR and both issuance variables. Given the strong positive association among the accrual measures, PPIR, and the issuance variables, the return 12 Size-adjusted returns are computed as the difference between the raw return as reported on CRSP and the size-matched return reported on the CRSP market capitalization decile return file crsp.ermport1. 13 Consistent with DT, I begin the future return series in July of year t+1 for all firms, regardless of the month in year t that their fiscal year-ends. Results are qualitatively identical when only firms with December fiscal year-ends are included. If a monthly return is missing during a return period because there is no price on CRSP a code of.b in the WRDS SAS data set, the return is set to zero. 14 Prior accrual studies have primarily defined accruals as a linear difference, scaled by average total assets e.g., NOA t NOA t 1, divided by average total assets. To be consistent with these studies, I winsorize linear difference variables, although results are not overly sensitive to this treatment.

9 Intangible Returns, Accruals, and Return Reversal 1355 TABLE 1 Summary Statistics BM ShrsOut Price ME TACC WCACC iss t 1,t Ret For All 61,241 Firms Mean , Std. Dev , Q Q Q For 1968: 633 Firms Mean Std. Dev , Min Max , For 1977: 1,179 Firms Mean Std. Dev , Min Max , For 1978: 1,457 Firms Mean Std. Dev , Min Max , For 2000: 1,811 Firms Mean , Std. Dev , Min Max , , This table reports the mean, standard deviation, lower quartile, median, upper quartile, minimum, and maximum for the primary sample of firm-year observations for firms with fiscal years ending t 1/31/67 and 12/31/06. BM is the book-tomarket ratio computed as book value of common equity as of fiscal year-end scaled by market value of equity as of fiscal year-end data25 data199. ShrsOut is equal to the number of shares outstanding in millions. Price is equal to the price per share. ME is equal to market value of equity. TACC is equal to total accruals computed as log NOA t / NOA t 1. WCACC is equal to working capital accruals computed as log WC t / WC t 1. WC t is computed as current assets data4 minus cash and short-term investments data1 minus current liabilities data5 plus debt in current liabilities data34 at fiscal year-end t. Specifically, working capital equals data4 data1 t data5 data34 t. iss t 1,t is a composite equity issuance variable computed as the difference between log change in the market value of equity between fiscal year-ends t 1 and t and the log market return inclusive of dividends between fiscal year-ends t 1 and t. Specifically, iss t 1,t = log ME t / ME t 1 Ret m t 1,t. Ret equals the buy-and-hold return, inclusive of distributions, between July 1 of year t+1 to June 30 of year t+2. See the Appendix for the Compustat data items that comprise book value of common equity and NOA. ME, ShrsOut, and Price are per CRSP and as of last day of trading in June, t+1. BM, TACC, WCACC, and iss t 1,t are of fiscal year-end t and are per the merged CRSP/Compustat database. reversal that characterizes the accrual anomaly may be due to investor reaction to information not measured by accruals, but correlated with the recognition of accruals. Regression Results The null hypothesis of this study is that future period returns are unrelated to realized accounting values such as accruals, consistent with the tenets of market efficiency and risk-neutral

10 1356 Resutek TABLE 2 Full Sample Correlation Statistics bm t PPIR iss t 5,t 1 iss t 1,t N_TACC TACC WCACC Ret bm t PPIR iss t 5,t iss t 1,t N_TACC TACC WCACC Ret Correlations significant at less than 0.01 level are bolded. This table reports full sample univariate correlations for log-linear return components of the book-to-market ratio with Pearson Spearman coefficients reported above below the diagonal. Sample includes firms with fiscal year-ends between 1/31/67 and 12/31/06. bm t equals the log book-to-market ratio as of fiscal year-end t with book value of common equity and market value of equity as of fiscal year-end t. PPIR is equal to prior period intangible log return, defined as the residual from annual regressions of log market return, Ret m t 5,t 1,oniss t 5,t 1, log B t 1 / B t 5, and bm t 5. iss t,t is a composite equity issuance variable computed as the difference between log change in the market value of equity between fiscal year-ends t and t and the log market return inclusive of dividends between fiscal year-ends t and t. N_TACC is equal to log B t / NOA t / B t 1 / NOA t 1. TACC is equal to total accruals computed as log NOA t / NOA t 1. WCACC is equal to working capital accruals computed as log WCACC t / WCACC t 1. Ret equals the buy-and-hold return, inclusive of distributions, between July 1 of year t+1 to June 30 of year t+2. See the Appendix for the Compustat data items that comprise book value of common equity and NOA. investors properly incorporating the pricing implications of accruals into price. However, if future period returns are significantly related to current period accruals, and this relation is not explained by risk or other correlated corporate activities such as equity issuance, then the null hypothesis can be rejected. To examine whether accruals are negatively associated with future returns after prior period intangible returns are controlled, I perform incremental regressions that reconcile my results to prior accrual literature and link the accrual anomaly to the value/growth anomaly. I begin by establishing the negative relation between current period accruals and future returns. I then introduce other tangible and intangible components of total market return over the current period. These variables complete the one-year decomposition of the bm t and largely confirm inferences from prior accrual literature. Finally, I complete the five-year decomposition of bm t and provide evidence that future return reversal is not associated with current period accruals, but rather the intangible returns that precede accrual recognition. Further, the five-year decomposition of bm t provides evidence that the accrual anomaly is a function of the value/growth anomaly. Table 3 reflects the results of Fama-MacBeth regressions of raw monthly returns on components of log book-to-market. 15 Regression 1 of Table 3 shows that, in a univariate specification, total accruals TACC are strongly negatively associated with future returns, consistent with prior studies. Regression 2 shows that the negative relation holds after controlling for non-accrual growth N_TACC. Regression 3 shows that, even after controlling for equity issuance and nonaccrual growth, there is still a strong negative association between TACC and future returns. Regression 4 includes current period intangible returns CPIR as an additional explanatory 15 Inferences from empirical analysis are qualitatively the same if size-adjusted returns are examined.

11 TABLE 3 Regressions of Future Monthly Returns on Log Return Components of Current Book-to-Market Model Int. TACC N_TACC iss t 1,t CPIR PPIR iss t 5,t 1 PPBG bm t *** 0.598*** *** 0.765*** *** 0.398** *** *** 0.399** *** 0.317*** *** 0.386** *** 0.338*** 0.390*** *** 0.374** *** 0.330*** 0.372*** *** 0.350** *** 0.318*** 0.364*** * *** *** 1.34 *** 0.337*** 0.284*** 0.376*** * *** ** 1.566*** 0.336*** 0.268*** *, **, *** Refers to significance at the 10 percent, 5 percent, and 1 percent levels, two-tailed, respectively. This table reports the results of a set of Fama-MacBeth regressions of monthly returns on components of a log-linear return decomposition of the book-to-market ratio at fiscal year-end t. The monthly return series is between July of year t+1 and June of year t+2 and t is between 1/31/67 and 12/31/06. All coefficients are multiplied by 100. t-statistics are reported in parentheses. bm t 5 is the log book-to-market ratio at fiscal year-end t 5. TACC is equal to total accruals computed as log NOA t / NOA t 1. N_TACC is equal to log B t / NOA t / B t 1 / NOA t 1. iss t,t is a composite equity issuance variable computed as the difference between log change in the market value of equity between fiscal year-ends t and t and the log market return inclusive of dividends between fiscal year-ends t and t. Specifically, iss t 1,t = log ME t / ME t 1 Ret m t 1,t. CPIR is the current period intangible return, defined as the residual from annual regressions of log market return, Ret m t 1,t,onbm t 1, TACC, N_TACC, andiss t 1,t. PPIR is the prior period intangible return, defined as the residual from annual regressions of log market return, Ret m t 5,t 1,onbm t 5, log B t 1 / B t 5, and iss t 5,t 1. PPBG is the prior period log book growth computed as log B t 1 log B t 5. See Appendix for the Compustat data items that comprise book value of common equity and NOA. Intangible Returns, Accruals, and Return Reversal 1357

12 1358 Resutek variable. 16 The significant negative relation between future period returns and CPIR suggests that, in expectation, current period returns are not a strict function of current period changes in book value. This finding is important because many accrual studies that adopt the Mishkin model implicitly assume that, in expectation, only earnings and earnings components explain future returns. As Kraft et al show, to the extent this assumption is incorrect and the omitted explanatory variable is mispriced, incorrect inferences can result regarding the relation between future period returns and current period accruals. 17 Subsequent specifications 5 7 introduce tangible return components from the four-year period prior to accrual recognition. Specifically, regressions 5 7 introduce prior period equity issuance iss t 5,t 1, prior period log book growth PPBG, and long-horizon expectations bm t 5. Inferences on these additional explanatory variables are consistent with those of DT. More importantly, even with the current and prior period log returns controlled, there is still an economically and statistically strong negative relation between future returns and current period accruals. However, once PPIR is introduced as an explanatory variable in regression 8, completing the log-linear return decomposition of bm t, accruals are no longer associated with future returns at conventional levels of significance. Building off regression 4, which shows that TACC are associated with future returns after current period tangible and intangible returns are controlled, regression 9 adds PPIR as a control variable. Regression 9 shows that, after PPIR is controlled, TACC is unrelated with future returns. This finding implies that PPIR is a powerful control variable that can distort inferences of current period accounting growth metrics and future returns when omitted. IV. ALTERNATIVE VARIABLE SPECIFICATIONS AND ROBUSTNESS TESTS Alternative Accrual Definitions Table 4 examines other measures of accruals to determine whether the exclusion of prior period intangible returns in prior studies explains the relation between accruals and future returns. The first set of regressions in Table 4, Panel A examines the relation between the components of total accruals suggested by Richardson et al and future returns. Richardson et al linearly decompose total accruals into three parts: change in sales Sales, change in asset turnover ATO, and an interaction between Sales and ATO. Using a log-linear decomposition avoids the complicating interaction term, yielding log total accruals TACC as the log sum of log Sales t 1,t and log ATO t 1,t. Consistent with the intuition of Richardson et al. 2006, Table 4, Panel A shows a strong statistical and economic relation between both log Sales t 1,t and log ATO t 1,t and future returns. 18 Subsequent controls for current and prior period log return components in regressions 2 and 3 show that log ATO t 1,t is still significantly associated with future returns, although log Sales t 1,t becomes insignificant with the introduction of iss t 1,t. 19 However, once PPIR is added as an explanatory variable in regression 4, completing the log-linear decomposition of bm t, neither accrual component is associated with future returns. 16 Since some firms used to compute CPIR do not have a full series of 12 future monthly returns, the coefficients on TACC, N_TACC, and iss t 1,t change slightly with the introduction of CPIR. 17 Alternatively, to the extent that the omitted variable is correctly priced, but the expected return model is incorrectly specified, incorrect inferences can also result. 18 Richardson et al do not directly examine the relation between future returns and accruals. However, the negative relation between future returns and accrual components reflected in Table 4, Panel A is consistent with the persistence coefficients reported in their study. 19 The lack of a significant relation between the growth component of accruals, log Sales t 1,t, and future returns is consistent with the results of Daniel and Titman 2006.

13 TABLE 4 Regressions of Future Monthly Returns on Log Return Components of Current Book-to Market: Alternative Accrual Specifications Panel A: Total Accrual Components Model Int. log Sales t 1,t log ATO t 1,t N_TACC iss t 1,t CPIR d PPIR iss t 5,t 1 PPBG bm t *** 0.729*** 0.717*** *** ** *** 0.334*** *** ** *** 0.339*** 0.361*** * *** *** 1.323*** 0.358*** 0.290*** 0.373*** * *** ** 1.564*** 0.353*** 0.273*** Panel B: Working Capital Accruals Model Int. WCACC N_WCACC iss t 1,t CPIR w PPIR iss t 5,t 1 PPBG bm t *** 0.547*** 0.300* *** *** 0.298** *** *** 0.305** 0.319** *** ** 1.527*** 0.327*** 0.285*** 0.336** *** ** 1.780*** 0.321*** 0.277*** *, **, *** Refers to significance at the 10 percent, 5 percent, and 1 percent levels, two-tailed, respectively. This table reports the results of a set of Fama-MacBeth regressions of monthly returns on components of a log-linear return decomposition of the book-to-market ratio at fiscal (continued on next page) Intangible Returns, Accruals, and Return Reversal 1359

14 year-end t. The monthly return series is between July of year t+1 and June of year t+2 and t is between 1/31/67 and 12/31/06. All coefficients are multiplied by 100. t-statistics are reported in parentheses. log Sales t 1,t is log change in sales data12 for fiscal period t 1 to t. log ATO t 1,t is log change in asset turnover for fiscal period t 1 to t where ATO t is defined as NOA t / Sales t. N_TACC is equal total non-accrual growth for the fiscal year t 1 to t. N_WCACC is equal to working capital non-accrual growth for the fiscal year t 1 to t defined as log B t / WC t / B t 1 / WC t 1. CPIR d is the current period intangible return defined as the residual from annual regressions of log market return, Ret m t 1,t,onbm t 1, log Sales t 1,t, log ATO t 1,t, N_TACC, iss t 1,t. CPIR w is the current period intangible return defined as the residual from annual regressions of log market return, Ret m t 1,t,on bm t 1, WCACC, N_WCACC, and iss t 1,t. TACC, WCACC, iss t,t, PPIR, PPBG, and bm t 5 are the same as defined in prior tables. See Appendix for Compustat data items comprising book value of common equity and NOA Resutek

15 Intangible Returns, Accruals, and Return Reversal 1361 Panel B of Table 4 examines working capital accruals. Due to the logarithmic nature of this variable, I omit depreciation from my working capital accrual variable. This omission will lead to a positive working capital accruals on average see Table 1, but should not affect regression inferences. 20 Regression 1 shows the telltale negative relation between WCACC and future returns. However, this relation becomes insignificant once iss t 1,t is introduced into the regression specification. Regressions 3 5 reinforce the validity of the relations noted in prior tables; that is, there is a strong negative relation between equity issuance and intangible returns with future returns, while the accrual component of earnings is unrelated with future returns. The relation between PPIR and current period accruals dovetails nicely into the intuition of prior studies that explain the accrual anomaly as a function of firm growth and investor misunderstanding of the diminishing returns on investment that follow firm growth. However, the evidence in this study suggests a different interpretation of the relation between future returns and current period accruals than that proposed in prior studies. While investors may misunderstand the diminishing returns on investment, this misunderstanding is not caused by the information contained in accruals or other accounting performance measures. Rather, the empirical evidence of this study suggests that the return reversal that prior studies have attributed to accounting measures of growth appears to be due to growth not captured by the financial accounting system, denoted as intangible returns. Linear Accrual Variable Prior literature has defined current period accruals as a linear difference scaled by assets. For example, total accruals have been defined as the one-year change in net operating assets scaled by beginning period assets. TACC linear = NOA t NOA t 1. TA t 1 Simple algebra shows that TACC linear can be expressed as the product of a beginning stock measure and a one-year growth measure: TACC linear = NOA t 1 TA t 1 NOA t 1. 7 NOA t 1 Equation 7 is useful because it shows that TACC linear is similar to the log-linear accrual variable TACC used in this study. 21 NOA Given that t 1 TA t 1 is positive for all observations in this study, TACC linear is a monotonic transformation of one-year growth in NOA. Thus, TACC linear should capture the same underlying economic dynamics as TACC, albeit it in a different scale. TACC linear captures growth on a linear basis, whereas TACC captures growth on a log-linear basis. This difference will cause TACC linear values to be more extreme than TACC values for observations with the most extreme positive NOA growth. Accordingly, differences in the regression coefficients of TACC linear and TACC may be due to differences in their respective econometric properties, not the economic dynamics they capture. Table 5 examines whether the primary results of Table 3 are sensitive to accrual measurement. Consistent with the intuition used in prior literature to explain the accrual anomaly, regression 1 establishes that the component of TACC linear responsible for the strong negative relation with future returns is the one-year growth in NOA component. Regression 2 directly examines the sensitivity of the results from Table 3 to the accrual definition used by regressing future monthly 20 Working capital accruals that include depreciation tend to have negative values, which are undefined in log form. 21 Empirically, the Pearson and Spearman correlations between TACC and TACC linear are 0.85 and 0.98.