Accrued Earnings and Growth: Implications for Earnings Persistence and Market Mispricing

Transcription

1 Accrued Earnings and Growth: Implications for Earnings Persistence and Market Mispricing by Patricia M. Fairfield a Scott Whisenant b Teri Lombardi Yohn a November 2001 Corresponding author Teri Lombardi Yohn The Robert Emmett McDonough School of Business Georgetown University Washington, DC (202) a Georgetown University, The Robert Emmett McDonough School of Business b The University of Houston, C. T. Bauer College of Business We appreciate the comments of Charles Lee, Richard Sloan, Srinivasan Sankaraguruswamy, Dennis Chambers, Sudipta Basu, Paul Hribar, Michael Mosebach, Ken Cavalluzzo, Dan Collins, Bill Baber, Son-Hyon Kang, Bala Dahran, Agnes Cheng, Stephen Zeff, and Tom Stober. We would especially like to thank Stephen Penman, James Ohlson, and Prem Jain for their contributions. This study has also benefitted from the comments of workshop participants at Emory University, Texas Christian University, The University of Houston, Rice University, George Washington University, The University of Arkansas, Hong Kong University of Science and Technology, and the 2001 American Accounting Association annual meetings. This research was supported in part by a Steers Faculty Research Fellowship awarded by the McDonough School of Business.

2 Accrued Earnings and Growth: Implications for Earnings Persistence and Market Mispricing ABSTRACT: An important goal of accounting research is to provide evidence that improves the analysis of financial statements for predicting future profitability. Research (Sloan 1996; Xie 2001) has found that (1) the persistence of earnings performance depends on the proportions of the cash and accrual components and that (2) a market inefficiency results from the failure of investors to fully appreciate the implications of cash flows and accruals for future earnings performance. In this study we investigate whether these results with respect to accruals can be generalized to another form of growth in net operating assets. We find that growth in long-term net operating assets, like accruals, has a negative association with one-year-ahead return on assets. We also find that the negative associations of both forms of growth (accruals and growth in longterm net operating assets) to one-year-ahead return on assets are attributable to the effect of growth on the denominator of return on assets. Furthermore, we find that the apparent market mispricing of accruals applies to growth in long-term net operating assets and that the severity of the mispricing does not significantly differ between the components of growth. Thus, the results suggest that the accrual anomaly documented in Sloan (1996) is a subset of a larger anomaly with respect to a general market mispricing of growth in net operating assets. Keywords: Accrued Earnings, Growth, Earnings Quality, Financial Statement Analysis, Market Mispricing

3 Accrued Earnings and Growth: Implications for Earnings Persistence and Market Mispricing I. INTRODUCTION Research on financial statement analysis has identified factors that aid in predicting profitability. For example, Freeman, Ohlson, and Penman (1982) documents regression toward the mean in return on equity. Other research has shown that financial statement components and ratios provide information about future earnings changes (Ou and Penman 1989; Ou 1990; and Abarbanell and Bushee 1997) and about future return on assets (Fairfield and Yohn 2001). 1 In addition, research has examined whether information in historical financial statements appears to be appropriately reflected in market values of equity. Research has documented market mispricing based on certain fundamental descriptors such as firm size (Fama and French 1992), recent earnings surprises (Bernard and Thomas 1989), and past sales growth (Lakonishok, Shleifer, and Vishny 1994). A recent notable contribution to both streams of literature is the finding of differential persistence of the accrual and cash flow components of earnings performance and the mispricing of the accrual and cash flow components (Sloan 1996; Collins and Hribar 2000; and Xie 2001). Sloan (1996) documents that the accrual component of earnings is less persistent than the cash flow component and concludes that investors fail to fully appreciate the differing properties of accruals and cash flows. In this study, we investigate whether the results regarding the lower persistence and market mispricing of accruals can be extended to apply to long-term growth in net operating assets and whether the lower persistence of accruals is due to the differential impact of growth in net operating assets relative to cash flows on the denominator of the earnings performance measure. Motivating this investigation is the observation that earnings performance, as defined in Sloan (1996), is operating income divided by contemporaneous average total assets which converts operating income into return on assets (ROA). In addition, accruals is defined as growth in operating working capital less depreciation and amortization 1 Ou and Penman (1989) reports a mechanical process where an array of financial statement ratios predicts future changes in earnings; Ou (1990) finds that variables other than earnings that are reported in financial statements predict one-year-ahead changes in earnings per share; Abarbanell and Bushee (1997) finds that financial statement ratios used by financial analysts predict future changes in earnings; and Fairfield and Yohn (2001) documents that changes in asset turnover are leading indicators of changes in future return on assets.

4 2 expense. Accruals, therefore, reflects not only a component of operating income in the numerator of current ROA, but also a component of growth in net operating assets which impacts average total assets in the denominator of one-year-ahead ROA. We argue that accruals, as a component of growth in net operating assets, is more highly correlated with average total assets than cash flows from operations. We predict that the documented lower persistence of accruals versus cash flows is attributable to the differences in correlations between the two components and the denominator of one-year-ahead ROA. An alternative interpretation of the differential persistence of accruals is that, conditioned on current ROA, accruals is negatively associated with one-year-ahead ROA. We posit that the documented negative association between accruals and one-year-ahead ROA, is attributable to the disproportionate impact of growth in net operating assets relative to cash flows on the denominator of the ratio. We, therefore, hypothesize a negative association between, not only accruals, but also growth in long-term net operating assets and one-year-ahead ROA. In addition, we hypothesize that the negative association will not be evident when the deflator of the earnings performance measure is lagged, rather than contemporaneous, average total assets. The results are consistent with our hypotheses. Specifically, we find that, conditioned on current ROA, growth in long-term net operating assets correlates negatively with one-year-ahead ROA. Furthermore, we find no differences between accruals and growth in long-term in net operating assets in their implications for one-year-ahead ROA. We also find that the negative associations between oneyear-ahead ROA and both forms of growth in net operating assets (accruals and growth in long-term net operating assets) are due to the impact of growth on the denominator of ROA rather than due to their implications for one-year-ahead operating income. These results are important for financial statement analysis. Our findings suggest that current ROA and current growth in net operating assets are associated with one-year-ahead ROA. 2 The finding is 2 We note that our results do not imply that accruals cannot be used to improve predictions of either future ROA or future operating income. However, we simply conclude that the evidence does not demonstrate a primary role for current accruals deflated by assets in explaining either one-year-ahead ROA or earnings. Nevertheless, we believe some other transformation of accruals, or total balance sheet growth, might well have incremental explanatory power for ROA and/or earnings (see, for example, Fairfield and Yohn 2001).

5 3 consistent with Ohlson (1995) and Feltham and Ohlson (1995) valuation models that suggest that profitability and growth are primary drivers for valuation. Our results suggest that, on average, no further explanatory power is provided by disaggregating current ROA into its accrual and cash components. The results are also important for future research. Researchers often use the terms earnings, ROA, and profitability interchangeably. 3 Our results demonstrate that the association between growth and one-yearahead ROA is not equivalent to the association between growth and one-year-ahead operating income. The choice of deflator is, therefore, important in the interpretation of results regarding the persistence of accruals for future income versus future ROA. Given that, conditioned on current ROA, there are equivalent associations between accruals and growth in long-term net operating assets and one-year-ahead ROA, we also investigate whether the market inefficiency with respect to accruals documented in Sloan (1996) extends to growth in long-term net operating assets. Using the Mishkin (1983) test, we test whether there is an inconsistency between the predictive ability of growth in long-term net operating assets for one-year-ahead ROA and the weight on growth in long-term net operating assets implicit in stock prices. We find that investors appear to overweight the implications of growth in long-term net operating assets for one-year-ahead ROA in a manner similar to the market mispricing of accruals. The two mispricing effects appear to be independent but statistically equivalent. The results suggest, therefore, that the apparent market mispricing of accruals can be generalized to apply to a single fundamental variable, growth in net operating assets. Collins and Hribar (2000, page 122) note that understanding what causes the market to fail to fully impound the future earnings implications of current accrual signals is a perplexing question that deserves further investigation. Our results make the question no less perplexing. Yet, we find that the accrual mispricing appears to be a general failure of the market to fully impound the value implications of growth in net operating assets.. The results suggest that the accrual anomaly documented in Sloan (1996) is a subset of a larger anomaly with respect to a general market mispricing of growth in net operating assets. 3 Sloan (1996) uses the term earnings performance in his study; however, he labels his variable Earnings in the analysis. Subsequent studies often fail to distinguish between earnings and ROA in summarizing the related literature.

6 4 The next section of the paper examines the associations between accruals, growth, and one-yearahead ROA. Section III provides a description of the variables used in the analysis. Section IV provides evidence on the association between one-year-ahead ROA and accruals and growth. Section V examines the effect of the denominator on the association between accruals, growth and one-year-ahead ROA, section VI examines the pricing anomaly with respect to accruals and growth in long-term net operating assets, and section VII presents the conclusions. II. ACCRUALS, GROWTH AND FUTURE PROFITABILITY Sloan (1996) finds that the accrual component of operating income is less persistent than the cash component for explaining one-year-ahead ROA. We note, however, that while accruals, as defined in Sloan (1996), 4 reflects a component of reported operating income, accruals also reflects growth in net operating assets on the balance sheet. Specifically, the accrual component of operating income (ACC) is defined as growth in operating working capital (GrWC) less depreciation and amortization expense (DEPAM): ACC t = GrWC t - DEPAM t Thus, accruals reflects a component of operating income as well as a component of growth in net operating assets. This is particularly relevant to the interpretation of the results in Sloan (1996) given the definition of earnings performance. Specifically, the dependent variable in the analysis is one-year-ahead ROA, defined as: 4 The goal of accrual accounting is to record value generated (revenue) or given up (expense) by operations when earned or incurred instead of when cash is received or paid. The outcome is to distinguish income from cash flows from operations. It is worth noting that the accrual process as described in SFAS No. 95 to reconcile income to cash from operations involves deferrals of past operating cash receipts and payments, accruals of expected future operating cash receipts and payments, and the effects of all items whose cash effects were previously classified as investing or financing cash flows (e.g., depreciation). Of these three types of reconciling items, only one is an accrual (as defined in SFAC No. 6). Nevertheless, to be consistent with prior literature, we term the difference between operating income and cash flows from operations to be accruals.

7 5, however, the independent variables are each measured and deflated by variables at time t. Therefore, one-year-ahead ROA is affected by growth in total assets in years t and t+1 in the denominator of the ratio. Unless there is perfect negative correlation in growth in assets, growth in total assets in year t will inflate the denominator of one-year-ahead ROA. 5 We argue that the differential persistence between cash flows and accruals for one-year-ahead ROA is attributable to their differential correlations with total assets in the denominator of the ratio. While cash flows and accruals both potentially increase total assets, a dollar of cash flows and a dollar of accruals are likely to have disproportionate impacts on total assets. The use of cash flows from operations for financing activities such as dividends and debt payments suggests that a dollar of cash flows from operations will increase total assets by, on average, less than a dollar. In contrast, because accruals reflects growth in current operating assets net of current operating liabilities, a dollar of accruals will increase total assets by, on average, more than a dollar. 6 Thus, we argue that accruals has a disproportionate impact on total assets in the denominator of one-year-ahead ROA. An alternative interpretation of the differential persistence of cash flows and accruals is that, conditioned on current ROA, accruals are negatively associated with one-year-ahead ROA. If the negative association of accruals with one-year-ahead ROA, conditioned on current ROA, is attributable to the disproportionate impact of growth in net operating assets relative to cash flows on the denominator of the ratio, a negative association should exist between not only accruals and one-year-ahead ROA but also growth in long-term net operating assets and one-year-ahead ROA. Growth in long-term net operating 5 We note that this effect will be more pronounced if there is serial correlation in growth in net operating assets. 6 This is confirmed by empirical evidence. In a regression of the change in total assets in year t on cash flows and accruals (all deflated by average total assets in year t), the coefficient of cash flows is 0.34 while the coefficient of accruals is The coefficients are significantly different from each other. When accruals is disaggregated into growth in operating working capital and depreciation expense, the coefficient of cash flows is 0.32, the coefficient of growth in operating working capital is 1.33, and the coefficient of depreciation is The coefficients on cash flows and accruals are reliably different from zero, while the coefficient of depreciation is not.

8 6 assets is not a component of reported income, but is a component of growth in net operating assets and should also have a disproportionate impact on average total assets relative to cash flows. As noted above, accruals reflects growth in current net operating assets. We note that total growth in net operating assets (GrNOA), which is defined as: GrNOA t = NOA t - NOA t-1, can be disaggregated as follows: GrNOA t = GrWC t - DEPAM t + GrLTNOA t.. GrLTNOA is defined as growth in gross property, plant and equipment and other long-term net operating assets. Growth in net operating assets can, therefore, be disaggregated into growth in the form of accruals (ACC) and growth in long-term net operating assets (GrLTNOA) as follows: 7 GrNOA t = ACC t + GrLTNOA t. We hypothesize that there will be a negative association between one-year-ahead ROA and accruals as well as growth in long-term net operating assets. Thus, our first hypothesis, in alternative form, is: Hypothesis 1: Conditioned on current ROA, growth in long-term net operating assets is negatively associated with one-year-ahead ROA. We also argue that if the negative associations between the components of growth in net operating assets and one-year-ahead ROA are due exclusively to the impact of growth on the denominator of ROA, the negative associations will not be evident when the deflator is the same for the dependent and explanatory variables. By using the same deflator for all the variables, we obtain the associations between the components of growth in net operating assets and one-year-ahead operating income. We hypothesize, therefore, that both components of growth in net operating assets will exhibit non-negative associations with one-year-ahead operating income. Thus, our second hypothesis, in alternative form, is: Hypothesis 2: The associations between the components of growth in net operating assets (accruals and growth in long-term net operating assets) and one-year-ahead operating income will be non-negative. 7 This is consistent with disaggregations of growth in net operating assets in prior research where growth in long-term net operating assets represents the net investment activities into long-term operating assets (Nissim and Penman 2001; Penman 2000).

9 7 III. VARIABLE DEFINITIONS Our sample consists of firms with required financial statement and stock price data for the 30 year period The purpose of using this sample period is to ensure replication of the Sloan (1996) results. Firms in the financial services industry (DNUM ) are removed since those firms typically do not have the required data to calculate accruals. The financial disclosure data are taken from the annual database of Standard & Poor s 1999 COMPUSTAT Industrial and Research files (Primary, Supplemental, Tertiary; Full-Coverage; and Industrial Research Files). The stock price and share data are obtained from the Center for Research in Security Prices (CRSP) 1999 monthly stock price database. The sample period includes only four reporting periods that have Statement Financial Accounting Standard No. 95 cash flow data from which one could directly estimate the accrual and cash flow components of operating income. Therefore, an alternative method is required to estimate accruals and cash flows from operations. Sloan (1996) uses a balance sheet approach that relies on the articulation between changes in working capital and accruals. To ensure comparability, we use the indirect method. We eliminate from the sample those observations without sufficient financial disclosures to compute operating income, accruals, cash flows, or net operating assets. There are 35,886 firm-years with sufficient stock price data and financial disclosure data to estimate the financial statement variables. Since the indirect method is used to estimate accruals, we also eliminate observations that would most likely induce noise into the measurement of accruals. We eliminate 2,806 firm-year observations impacted by one or more of the following since each leads to increased noise in the measurement of either accruals or net operating assets. That is, we eliminate observations where (1) working capital components are estimated, (2) voluntary accounting changes made by managers impact either working 8 Sloan (1996) employs financial data for the 30 years from 1962 to However, we drop observations from 1962 from our sample since the required data reduce the sample size for that year to below a reasonable sample size on which to run annual cross-sectional regressions. Therefore, we add one additional year, 1992, to our sample period which yields a 30-year sample period as in Sloan (1996). The conclusions of this study are unchanged if observations from 1992 are removed.

10 8 capital or operating income, or (3) the recorded amount of goodwill increases from year to year. 9 The sample selection process results in a final sample of 33,080 observations with available stock price and financial statement data in the prior, current, and subsequent year. 10 Measurement of the Variables The explanatory variables in the regression analyses are current ROA, cash flows, accruals, and growth in long-term net operating assets. Consistent with Sloan (1996), ROA is defined as operating income deflated by contemporaneous average total assets. 11 Thus, ROA is defined as (although not shown, cross sectional firm subscripts throughout the study should be assumed):, 9 We first eliminate observations for which COMPUSTAT footnote codes suggest that certain classifications such as current assets, current liabilities, and debt due in one year, are estimated since the company reports an unclassified balance sheet (364 firm-years). In the second step, we eliminate observations where COMPUSTAT footnote codes suggest that an income statement item reflects an accounting change (2,059 firm-years) that is a voluntary change made by managers that impacts either operating income or working capital accounts. In the third step, we remove observations for which there was a one-year increase in goodwill so as to remove firms that had purchase acquisitions and paid in excess of fair value of net assets (597 firm-years). We eliminate observations with increases in year-to-year goodwill since a purchase business combination means that changes in the balance sheet accounts may not articulate with income statement accruals. We consider an increase to be a positive year-to-year change in goodwill (COMPUSTAT item 240) where the data item has a non-missing value. Since some firm-years are impacted by more than one of these potential measurement error variables, the net elimination of 2,806 firm-years results. Although our results on differences in financial statement variables (accruals and cash flows, and accruals and growth in long-term net operating assets) are robust to the inclusion of the 2,806 deleted firm-years, we delete each since the methodology measures the variables of interest with less measurement error than if included. 10 Studies document nonarticulation problems with the indirect (or balance sheet) approach to estimate the accrual component of earnings (e.g., see Drtina and Largay 1985; Bahnson et al. 1996; Collins and Hribar 1999). For example, Collins and Hribar (1999) document that merger and acquisition activities, foreign currency translations, and divestitures introduce significant measurement error into accruals estimated using the indirect (balance sheet) method. In our sample, using COMPUSTAT disclosures, we find approximately 15% report merger and acquisition activities, 15% report foreign currency translations, and approximately 10% report divestitures. Elimination of firm-years affected by one of the three types of transactions leading to potential measurement error in accruals eliminates approximately 12,000 firmyears. The results reported in the tables are qualitatively similar when those firms are eliminated. We report results without deleting the 12,000 firm-years so as to enhance comparability with the results shown in Sloan (1996) and to provide for reasonable sample sizes in our year-by-year regressions. However, we note that, as expected, if these observations are removed from the sample, and all variables in the data matrix are deflated by beginning of the period total assets, we find no evidence of differential persistence in accruals and cash flows. 11 We scale all variables by contemporaneous average total assets to provide direct comparisons with the results reported in Sloan (1996). Following Feltham and Ohlson (1995) and Penman (2000), we argue that scaling by beginning of period net operating assets yields more economically meaningful accounting descriptors. All conclusions of the paper are robust to this alternative deflator.

11 9 where OPINC t = operating income after depreciation and mortization (COMPUSTAT item 178) TA t = total assets (COMPUSTAT item 6). As in Sloan (1996), we use the indirect (balance sheet) method to estimate accruals (ACC). 12 Using the indirect method, accruals is defined as the net change in operating working capital accounts other than tax liabilities minus current period depreciation and amortization. 13 Therefore, we define accruals as: ACC t = GrWC t - DEPAM t, where: GrWC t = ( AR t + INV t + CAO t ) - ( AP t + CLO t ), and AR = change in accounts receivable (COMPUSTAT item 2), INV = change in inventories (COMPUSTAT item 3), CAO = change in other current assets (COMPUSTAT item 68), AP = change in accounts payable (COMPUSTAT item 70), and CLO = change in other current liabilities (COMPUSTAT item 72). We estimate cash flows from operations (CFO) by subtracting the estimate of accruals from operating income (OPINC) as follows : 12 Note that this method does not consider accruals attributable to non-operating income or any other items below operating income on the income statement. The measurement of accruals by the indirect method therefore suggests a caveat when comparing the results in Sloan (1996) to the results of studies using SFAS No. 95 data (e.g., Collins and Hribar 2000) where accruals are defined using income before extraordinary items and discontinued operations minus cash flows from operations taken from the statement of cash flows. The result of testing the persistence of such estimates of accruals is quite different when using the direct versus the indirect methods. Fairfield, Sweeney, and Yohn (1996) find that the weights to income statement items for the prediction of future return on equity decrease as its classification is lower on the income statement (i.e., going from operating income toward net income). In a related study, Barth, Cram, and Nelson (2001) find that the accruals not considered by the indirect method exhibit lower persistence to future cash flows from operations. Therefore, evidence of lower persistence in studies using the direct method may not reflect the same phenomenon as those using the indirect method (i.e., in either our study or Sloan 1996). Caution should be used in interpreting the results of studies investigating differences in the persistence of accrual versus cash basis earnings when different methods are used to estimate the accrual and cash components. 13 Short-term taxes are removed since operating income excludes tax expense. The income measure also excludes interest expense, suggesting that interest payable should also be excluded. Interest accruals arising from differences between periodic interest payments and recorded interest expense are recorded in the net book value of debt and are, therefore, excluded from the estimate of accruals. However, interest accruals arising due to timing differences at fiscal year-end are reported in COMPUSTAT in either debt in current liabilities or current liabilities-other. When reported in debt in current liabilities, they are appropriately excluded from accruals. When reported in current liabilitiesother, the interest accrual is incorrectly included in accruals. Our results are robust to the exclusion of other current liabilities excluding accrued expenses in the estimate of accruals.

12 10 CFO t = OPINC t - ACC t. 14 Both variables, ACC and CFO are deflated, as in Sloan (1996), by contemporaneous average total assets. As accruals is a component of net operating assets, we define the remaining component of net operating assets, growth in long-term net operating assets (GrLTNOA), as growth in net operating assets less accruals. Specifically, we define the second component of growth in net operating assets: GrLTNOA t = GrNOA t - ACC t, where growth in net operating assets (GrNOA) is defined as annual changes in net operating assets: GrNOA t = NOA t - NOA t We use net operating assets to be consistent with the income measure used in the numerator of ROA (operating income) and the use of growth in operating working capital in the accruals measure. Net operating assets (NOA) is defined as operating assets minus operating liabilities at year t. 16 Specifically, we define NOA as: NOA t = operating assets (OA t ) - operating liabilities (OL t ), where operating assets (OA) and operating liabilities (OL) are defined as: OA t = AR t + INV t + CAO t + PPE t + INT t + AO t, and OL t = AP t + CLO t + LO t, where AR = accounts receivable (COMPUSTAT item 2), INV = inventories (COMPUSTAT item 3), CAO = other current assets (COMPUSTAT item 68), PPE = net property, plant, and equipment (COMPUSTAT item 8), 14 We also considered the period 1988 to 1998, the period for which statement of cash flow data are available on COMPUSTAT. With this data, we can use the direct, instead of the indirect, method to calculate accruals and cash flows. Our primary results regarding the similar implications of accruals and growth in long-term net operating assets after controlling for current total ROA are robust for this data if we consider accruals to be those resulting from changes in operating working capital accounts as in Sloan (1996). 15 The results are robust to a specification that removes depreciation and amortization from ACC and includes it with GrLTNOA. 16 The definition of net operating assets is consistent with Nissim and Penman (2001) and Penman (2001). However, unlike Nissim and Penman (2001) and Penman (2001), we define income as operating income instead of comprehensive income. Therefore, we exclude income taxes payable and deferred taxes and investment tax credits as well as investments and advances under the equity method from our definition of operating assets and operating liabilities since the impact of each is classified on the income statement below our income measure.

13 11 INT = intangibles (COMPUSTAT item 33), AO = other long-term assets (COMPUSTAT item 69), AP = accounts payable (COMPUSTAT item 70), CLO = other current liabilities (COMPUSTAT item 72), and. LO = other long-term liabilities (COMPUSTAT item 75). Growth in long-term net operating assets is, like the other explanatory variables, scaled by contemporaneous average total assets. Descriptive Statistics Table 1 provides descriptive statistics on ROA, accruals, cash flows, total growth in net operating assets, growth in long-term net operating assets, and the components of accruals (growth in working capital and depreciation) in year t. The mean (median) ROA for the firms in the sample is 11.6% (11.1%). As documented in prior studies, the mean (median) value of accruals is (-0.026), suggesting that accruals, on average, decrease income. However, the mean (median) value for growth in working capital is (0.015), suggesting that accruals are income decreasing due to the deduction of depreciation and amortization expense. The mean (median) value of growth in long-term net operating assets is (0.07), while the mean (median) value of total growth in net operating assets is (0.058). Growth in both working capital and long-term net operating assets are, on average, positive. [Please place table 1 about here] Table 2 describes the results of sorting the sample into equal deciles based on accruals. The patterns for the data reported in panel A are consistent with those reported in Sloan (1996). Return on assets increases over the accrual deciles and cash flows correlate negatively with accruals. In panel B we show that growth in net operating assets increases monotonically over the accrual deciles. We also decompose accruals into its two components: growth in working capital and depreciation. The results show that growth in working capital increases monotonically over the accrual deciles while depreciation decreases over the accrual deciles. In addition, the table documents that growth rates in long-term net operating assets are stable across the deciles except for the extreme deciles of accruals. [Please place table 2 about here]

14 12 Table 3 reports correlations among one-year-ahead ROA, and current ROA, accruals, cash flows, total growth in net operating assets, growth in long-term net operating assets, and the components of accruals. As expected, accruals exhibits a lower correlation with one-year-ahead ROA than does cash flows. Both components of growth exhibit positive correlations with one-year-ahead ROA. Additionally, accruals and growth in long-term net operating assets are positively correlated suggesting that growth tends to occur simultaneously in both short-term and long-term net operating assets. [Please place table 3 about here] IV. EVIDENCE ON THE ASSOCIATIONS BETWEEN ACCRUALS AND GROWTH IN LONG- TERM NET OPERATING ASSETS, AND FUTURE ROA In this section, we investigate the associations among the components of growth and one-year-ahead ROA if conditioned on current ROA. We first investigate the association between current and one-yearahead ROA. Equation (1) is estimated, by year. 17 (1) Table 4 reports the results of estimation of equation (1). Since it is well documented that profitability measures have a strong stable component (Penman 1991), the high degree of association between current and one-year-ahead ROA is expected. The estimated coefficient of current ROA is 0.721, suggesting that ROA is slowly mean-reverting. In equation (2), current ROA is decomposed into accruals and cash flows: (2) The estimated coefficient of accruals is and cash flows from operations is A matched paired t-test of the coefficients, reported in panel B, suggests a significant difference (p-value < 0.01) between 17 As in Fama and McBeth (1973), we estimate annual regressions and report t-values based on the means and standard deviations of the parameter estimates obtained in the annual regressions. Similar to the supplemental tests presented in Sloan (1996), we perform the pooled analyses, and analyses by industry groups and by return on asset deciles. We also perform the analyses using the Belsley, Kuh, and Welsch (1980) procedure to control for outliers. The result that accruals and growth in long-term net operating assets are not differentially related to future ROA after controlling for total ROA is robust to the these sensitivity tests.

15 13 accruals and cash flows from operations in their implications for one-year-ahead ROA. Thus, the results of estimating equation (2) on our sample confirm the differential persistence of accruals and cash flows documented in Sloan (1996). [Please place table 4 here] Analysis of the Components of Growth in Net Operating Assets As reported in Sloan (1996), we find that accruals are less persistent than cash flows for one-yearahead ROA. We note, however, that accruals reflects not only a component of operating income but also a component of growth in net operating assets. We also note that growth in net operating assets disproportionately impacts total assets in the denominator of one-year-ahead ROA. If the negative association between accruals and one-year-ahead ROA is attributable to the impact of accruals as a component of growth on the denominator of the ratio, an equivalent association between growth in longterm net operating assets and one-year-ahead ROA is expected. To investigate this issue we first use a alternative, but equivalent, specification of equation(2), in which current ROA and accruals are included as explanatory variables: (3) Using this specification, the explanatory variable, ACC, captures the differential information in accruals versus total ROA. Table 5 shows the results of estimating equation (3). The results highlight that, conditioned on total current ROA, growth in the form of accruals is negatively associated with one-yearahead ROA. In equation (4), we add growth in long-term net operating assets to the regression: (4) We can test whether the coefficients of accruals and growth in long-term net operating assets are equivalent. This is important since accruals reflects a component of current ROA as well as being correlated with the denominator of one-year-ahead ROA. In contrast, growth in long-term net operating assets is not a component of operating income but is correlated with the denominator of one-year-ahead

16 14 ROA. Therefore, we test whether accruals (as a source of growth) differs from another source of growth in net operating assets in its implications for one-year-ahead ROA when conditioned on current ROA. In Table 5, we report the expected negative coefficient of accruals and also observe a significant negative coefficient of growth in long-term net operating assets. In panel B, we test whether accruals is equivalent to growth in long-term net operating assets in their implications for one-year-ahead ROA. We find no reliable difference between the coefficients of accruals and growth in long-term net operating assets. 18 These results support the notion that the negative association between one-year-ahead ROA and accruals is not unique to accruals. These results are important for financial statement analysis. They suggest that, conditioned on current ROA and current growth, distinguishing between growth in the form of accruals and growth in long-term net operating assets provides no further explanatory power for future ROA. Equivalently, conditioned on current ROA and current growth, distinguishing between the cash and accruals components of ROA provides no further explanatory power for future ROA. [Please place table 5 here] V. EVIDENCE ON THE EFFECT OF GROWTH IN THE DENOMINATOR OF ROA ON THE ASSOCIATIONS AMONG ACCRUALS AND GROWTH AND FUTURE ROA The results in the previous section suggest that one-year-ahead ROA is negatively associated with accruals and growth in long-term net operating assets if conditioned on current ROA. The evidence also shows that growth in the form of accruals does not differ from growth in long-term net operating assets in explaining one-year-ahead ROA. We predict that the negative associations between accruals and oneyear-ahead ROA as well as growth in long-term net operating assets and one-year-ahead ROA are likely to be attributable to the impact of growth on the denominator of one-year-ahead ROA. If the negative associations are due exclusively to the effect of growth on the denominator of ROA, the negative 18 We also performed the analysis within two-digit SIC codes. The coefficients of growth in long-term net operating assets, accruals, and cash flows differ across the industries. However, the main result that accruals and growth in longterm net operating assets do not differ in their implications for one-year-ahead ROA holds across the industries.

17 15 associations will not be evident if the same deflator is used for the dependent and explanatory variables. Using the same deflator for all the variables, we can investigate the associations between the components of growth in net operating assets and one-year-ahead operating income. We hypothesize, therefore, that accruals and growth in long-term net operating assets will exhibit non-negative associations to one-yearahead operating income. In Table 6 we report the results of tests on whether the reported differential persistence of accruals and cash flows is due to the effect of growth on the denominator of the dependent variable. We present the results of re-estimating equation (3) by scaling the dependent variable (one-year-ahead operating income) by lagged average total assets instead of contemporaneous average total assets. That is, we estimate the following:. (5) The goal is to remove the impact of growth on the denominator of ROA and to isolate the association between accruals and the numerator of the ratio (one-year-ahead operating income). Table 6 reports the results of estimating equation (5). We find that the coefficient of accruals is not significantly different from zero. Thus, the results suggest that the lower persistence of accruals relative to cash flows can be attributed to the differential impact of accruals relative to cash flows on the denominator of one-year-ahead ROA. We then include growth in long-term net operating assets as an explanatory variable in the regression:. (6) We find that the estimated coefficient of growth in long-term net operating assets is 0.030, and is significantly different from zero at the five percent level. That is, growth in net operating assets appears to be positively associated with one-year-ahead operating income. More importantly, neither component of growth in net operating assets exhibits a reliably negative coefficient, and cross-sectional comparisons of the coefficients demonstrate that they are statistically equivalent.

18 16 The results suggest that growth in net operating assets, whether in the form of accruals or growth in long-term net operating assets, has minimal explanatory power for one-year-ahead operating income. The results support the hypothesis that the negative association between one-year-ahead ROA and growth in net operating assets is attributable to the impact of the growth on the denominator of one-year-ahead ROA. 19 This is important for interpreting the results reported here and in prior studies. The results suggest that the documented lower persistence of accruals is not due to accruals being negatively associated with future operating income. Instead, the accruals and cash flows appear to be differentially correlated with the denominator of the ROA ratio which leads to the differential persistence documented in Sloan (1996). 20 This is important for research that examines accruals as a signal of earnings quality. Researchers should be cautious not to interpret equivalently the results from estimating regressions where earnings and ROA are dependent variables. Our results suggest that accruals has different implications for one-year-ahead ROA and operating income. [Please place table 6 about here] 19 We note that the associations between accruals and one-year-ahead ROA as well as growth in net operating assets and one-year-ahead ROA could be driven by the effect of growth in net operating assets (or total assets) in year t+1. The result is possible if net operating assets exhibit a positive serial correlation. We find that the Pearson correlation between growth in net operating assets at years t and t+1 is 0.14 (significant at the one percent level). Therefore, we also perform the analysis including growth in net operating assets (and total assets) in year t+1 deflated by average total assets in year t as an additional explanatory variable. The results, not reported in a table, suggest that after including growth in net operating assets (or total assets) in year t+1 in either equation (3) or (4), reliably negative coefficients of accruals and growth in long-term net operating assets remain. These results suggest that the negative association between growth in net operating assets in year t and ROA in year t+1 does not appear to be driven by growth in net operating assets (or total assets) in year t+1. Additionally, after including growth in net operating assets (or total assets) in year t+1 as an explanatory variable in equations (5) and (6), we find no evidence of reliably negative coefficients of accruals or growth in long-term net operating assets in year t. These results suggest that the associations between accruals and growth in long-term net operating assets in year t as well as ROA and operating income in year t+1 are not attributed to the effect of growth in net operating assets (or total assets) in year t Barth and Kallapur (1996) find that including a scale proxy as an independent variable is more effective than deflation at mitigating coefficient bias, even if the proxy is 95 percent correlated with the true scale factor. In fact deflation can worsen coefficient bias. That is, the coefficient of the variable that is more highly correlated with the scale will be more biased. We re-estimated equations (5) and (6) with the scale as an independent variable. The results are qualitatively similar: non-negative values of accruals and growth in long-term net operating assets and equivalency between the two in their implications to one-year-ahead operating income.

19 17 VI. EVIDENCE ON THE MARKET MISPRICING OF GROWTH Prior research (Sloan 1996; Xie 2001; DeFond and Park 2001) concludes that stock prices fail to reflect the different properties of accruals versus cash flows. Given the evidence that, conditioned on current ROA, growth in long-term net operating assets, like accruals, correlates negatively with one-yearahead ROA, we posit that the investor mispricing documented with respect to accruals applies also to growth in long-term net operating assets. Thus, our third hypothesis, in alternative form, is: Hypothesis 3: There is an inconsistency between the predictive ability of growth in long-term net operating assets with respect to one-year-ahead ROA and the weight on growth in long-term net operating assets implicit in stock prices. In table 7, we provide evidence on the mispricing of accruals and growth in long-term net operating assets. Mishkin (1983) develops a framework to test whether investors appear to rationally price variables previously known by investors. The test provides a statistical comparison between a measure of the apparent pricing by investors of both forms of growth (i.e., the valuation coefficient of the growth components) and a measure of the ability of the components of growth to predict one-year-ahead ROA (i.e., the forecasting coefficient of the growth components). If the valuation coefficient of either component of growth is significantly larger than the forecasting coefficient of the components, then the Mishkin (1983) test would suggest that investors appear to overweight the implications of growth. Specifically, we estimate the following system: where the forecasting equation is:, (7) and the valuation equation is:, (8) where BHAR t+1 is the size-adjusted abnormal stock return computed as the difference between a firm s annual buy-and-hold return and the annual buy-and-hold return for the same 12-month period on the market-capitalization-based portfolio decile to which the firm belongs.

20 18 Equation (7) estimates the forecasting coefficients ( s) on current ROA, accruals, and growth in longterm net operating assets for predicting one-year-ahead ROA. Equation (8) is the valuation equation that estimates the valuation coefficients ( *s) that the market appears to assign to current ROA, accruals, and growth in long-term net operating assets. As discussed in Mishkin (1983), we estimate equations (7) and (8) jointly using an iterative generalized nonlinear least squares estimation procedure, proceeding in two stages. In the first stage, we impose no constraints on s and *s and jointly estimate equations (7) and (8). To test whether the valuation coefficients ( *s) are significantly different from the offsetting forecasting coefficients ( s) obtained in the first stage, we estimate equations (7) and (8) jointly in the next stage after imposing the rational pricing constraints, * q = q, where q = 1 or Mishkin shows that the following likelihood ratio statistic is distributed asymptotically as 2 (q) under the null hypothesis that the market rationally prices one or more independent variable in equation (7) with respect to their implications for one-year-ahead ROA: where: 2n log (SSR c / SSR u ), q = number of rational pricing constraints, n = the number of observations, log = natural logarithm operator, SSR c = the sum of squared residuals from the constrained weighted system, and SSR u = the sum of squared residuals from the unconstrained weighted system. Panel A of table 7 reports the estimated coefficient for equations (7) and (8) obtained from the first stage. We reject rational pricing of accruals, consistent with the results reported in Sloan (1996). For accruals, the valuation coefficient ( * 2 = 0.069) is significantly larger than the forecasting coefficient ( 2 = ), suggesting that the market overweights accruals relative to its ability to predict one-year-ahead ROA. We also reject the rational pricing of growth in long-term net operating assets. For growth in longterm net operating assets, the valuation coefficient ( * 3 = 0.051) is significantly larger than the 21 Mishkin (1983) offers a description of the methodology in Appendix 2.2.

21 19 forecasting coefficient ( 3 = ), suggesting that the market overweights growth in long-term net operating assets relative to its ability to predict one-year-ahead ROA. The apparent overpricing of each component of growth in net operating assets is significant at any reasonable level. 22 To test whether the overpricing of one of the growth components is more severe, we jointly estimate equations (7) and (8), in the second stage, after imposing the rational pricing constraint on the two forms of growth (i.e., * 2 = * 3 and 2 = 3 ). We fail to reject this special case of the null hypothesis, and conclude that the market mispricing of growth in long-term net operating assets relative to its ability to predict one-year-ahead ROA is no different than the mispricing of accruals. [Please place table 7 about here] We also perform tests of rational pricing of accruals and growth in long-term net operating assets using a forecasting equation that includes operating income in year t+1 deflated by lagged average total assets as the dependent variable. The results of reestimating equations (7) and (8) are not reported in table format since the conclusions regarding the mispricing of accruals and growth in long-term net operating assets are robust. That is, we reject rational pricing of both accruals and growth in long-term net operating assets, suggesting that the market overweights accruals and growth in long-term net operating assets relative to their ability to predict one-year-ahead operating income. We note, however, that the forecasting coefficients on accruals and growth in net operating assets are positive for one-yearahead operating income. Therefore, while growth is positively associated with one-year-ahead operating income, the market appears to place higher valuations on the growth components than are suggested in the forecasting equations. We also test whether there is an inconsistency between the predictive abilities of current operating income, accruals, and growth in long-term net operating assets with respect to one-year-ahead operating 22 Although not reported in a table, we also perform a hedge-portfolio test on the accrual component of profitability and growth in long-term net operating assets. To form the portfolios, firms are ranked and assigned to deciles based on accruals and growth in long-term net operating assets. We form the hedge portfolios taking long positions in the lowest portfolios (i.e., low accrual or low growth in long-term net operating assets) and short positions in the highest portfolios for the year following the portfolio formation. The results of both hedge-portfolio tests provides support for the inferences from the Mishkin (1983) test that the market appears to overweight both components of growth in net operating assets