Estimating Abnormal Changes in Cash with Earnings Persistence and Mispricing Implications *

Transcription

1 Estimating Abnormal Changes in Cash with Earnings Persistence and Mispricing Implications * Jeff Zeyun Chen Philip B. Shane First Draft: 27 July 2010 This draft: 9 August 2010 *We appreciate helpful comments of University of Colorado workshop participants. Corresponding author. University of Colorado at Boulder, Leeds School of Business. Voice: (303) zeyun.chen@colorado.edu. University of Colorado at Boulder Leeds School of Business, and the University of Auckland Business School. Voice: phil.shane@colorado.edu or p.shane@auckland.ac.nz. 0

2 Estimating Abnormal Changes in Cash with Earnings Persistence and Mispricing Implications 1. Introduction This paper extends research investigating the persistence and market pricing of earnings components. Sloan [1996] separates earnings into cash and accrual components and finds that the cash component has greater persistence than the accrual component, investors inefficiently treat both components as having similar persistence characteristics and, therefore, the stock market underreacts to the persistence of cash earnings and overreacts to the persistence of accruals. Xie [2001] further disaggregates accrual earnings into discretionary and nondiscretionary components and finds that discretionary accruals drive the accrual anomaly discovered by Sloan. 1 Dechow, Richardson and Sloan [DRS 2008] more precisely define total accruals as the income effects of changes in non-cash operating asset and liability accounts and free cash flow as the difference between total earnings and accruals. Furthermore, DRS disaggregate total free cash flow into net changes in the firm s cash balance (including short-term investments and other financial assets) and net distributions to providers of capital. DRS find that the change in the cash balance has similar persistence characteristics as accruals and that the market similarly overreacts to the persistence of both accruals and changes in cash. Our paper further analyzes the primary DRS finding. We investigate the characteristics of changes in cash that create persistence characteristics similar to accruals. In particular, as explained in more detail below, we disaggregate changes in cash into a normal (precautionary) 1 Richardson et al. [2006] describe another approach to disaggregating total accruals, but their goal is to investigate whether accounting distortions or diminishing returns to new investment drive the accrual anomaly. Richardson et al. conclude that accounting distortions are primarily responsible for the accrual anomaly, these distortions exist in both their asset turnover and sales growth components of total accruals, and they cannot rule out the Fairfield et al. [2003] diminishing returns to investment explanation for the accrual anomaly. 1

3 component and an abnormal component that can be positive or negative and creates either excess or insufficient cash on hand. In addition, we extend Xie s [2001] decomposition of accruals into discretionary and non-discretionary components to include long-term accruals (Richardson et al. [2005]) and we follow the performance-based matching approach recommended by Kothari et al. [2005]. Following DRS, we apply the Mishkin [1983] approach in our evaluation of market efficiency. 2 We extend DRS by evaluating market efficiency with respect to the persistence of seven earnings components: total accruals, disaggregated into discretionary and nondiscretionary components; and free cash flow, disaggregated into distributions to debt holders, distributions to equity holders, and normal, insufficient and excessive changes in cash. Our disaggregation of changes in cash follows recent developments in finance literature estimating optimal cash balances (e.g., Opler et al. [1999]; Bates, Kahle and Stulz [BKS 2009]). DRS disaggregate the free cash flow component of earnings into three categories: the change in the cash balance, net distributions to debt holders, and net distributions to equity holders. 3 DRS find that the change in cash and distribution to debt holders components of free cash flow have similar persistence as total accruals and lower persistence than distributions to equity holders, and the market similarly overreacts to both accruals and the portion of earnings (and free cash flow) due to the change in cash. DRS infer that net cash distributions to equity holders drive Sloan s [1996] finding that cash earnings have greater persistence than accruals. DRS find no 2 Further extending DRS, in sensitivity tests, we evaluate market efficiency using an OLS approach advocated by Kraft, et al. [2006] and Shane and Brous [2001]. Also, to avoid the look-ahead bias inherent in pooled crosssectional regressions, our sensitivity analysis applies both Mishkin tests and OLS in a Fama-MacBeth [1973] yearby-year framework. Conclusions regarding market inefficiency with respect to the persistence characteristics of all four free cash flow components are robust; however, as in Kraft et al. [2006], conclusions regarding market inefficiency with respect to accruals are not robust. 3 Throughout the paper, references to changes in the cash balance include changes in marketable securities and any other financial assets. Like changes in cash, net distributions to debt and equity holders can be positive or negative; i.e., the firm s payments to debt holders and stockholders can exceed receipts due to new investments in the firm s debt and equity securities or vice versa. 2

4 evidence of market inefficiency with respect to the cash distributions to equity and debt holder components of free cash flow; however, they find that the market similarly overreacts to the persistence of accruals and the change in cash earnings components. DRS (p. 558) attribute the finding of similar market overreaction and lower persistence of accruals and changes in the cash balance to hubris concerning future investment opportunities. If managers and investors are overoptimistic about the investment opportunities of certain firms, these firms invest more capital and have less sustainable profitability. Our paper further investigates the DRS hubris interpretation of their results. To do this, we disaggregate the change in cash component of free cash flow into a normal part that moves the firm towards an estimated optimum level of cash holdings and an abnormal part that moves the firm away from the estimated optimum. We further disaggregate abnormal changes in the cash balance into those that create excess cash and those that create insufficient cash. The hubris hypothesis implies that excessive changes in the cash balance lack persistence and that the market overreacts to the persistence of this component of free cash flow. We also investigate the hypothesis that insufficient free cash flow leaves the firm in a vulnerable position with respect to the ability to quickly take advantage of positive net present value investments when they arise (Opler et al. [1999]). Thus, we divide the change in cash component of free cash flow into three parts and, relative to DRS, provide a more detailed evaluation of the persistence and pricing of the change in cash component of free cash flow. The three parts are: normal changes in cash that move the firm towards an estimated optimal equilibrium level of cash holdings; changes in cash creating excessive cash on hand; and changes in cash detracting from the optimal level and creating insufficient cash on hand. 3

5 To estimate the optimal change in cash on hand, we follow a line of finance literature extending from Keynes [1934], who introduces the notions of transactions costs and precautionary motives for holding cash, and Miller and Orr [1966] and Jensen and Meckling [1976] who, respectively, develop the notions of brokerage and inefficient investment costs associated with holding cash. Opler et al. [1999], BKS and others have extended the theory, conceptualizing and testing hypotheses and developing models to estimate an optimal amount of cash holdings and changes in cash holdings. 4 As described by BKS (p ), the models include proxies for four factors explaining the level and change in a company s cash balance. Firms manage their cash balance to: (1) avoid transactions costs associated with liquidating assets (financial and operating) to meet obligations or make investments (Miller and Orr [1966]); (2) avoid repatriation of foreign earnings that would be taxed at a higher rate (Folie, et al. [2007]); (3) avoid missing investment opportunities when financing costs are high and cash flows are risky or the correlation between operating income and investment opportunities is low (Han and Qiu [2007], Acharya et al. [2007]); and (4) extract rents from shareholders by entrenched managers of firms with high agency costs (Jensen and Meckling [1976], Jensen [1986]). For purposes of the exposition of our paper, we refer to the first three factors as precautionary and the fourth factor as agency-related. The agency-related factor refers to costs associated with inefficient investment and perquisites extracted by managers. We rely primarily on the BKS model to estimate optimal changes in cash for precautionary reasons. The residual of the model provides an estimate of 4 For example, see Keynes [1934], Chudson [1945], Miller and Orr [1966], Vogel and Maddala [1967], Jensen and Meckling [1976], Myers [1977], Myers and Majluf [1984], Baskin [1987], John [1993], Beltz and Frank [1996], Mulligan [1997], Harford [1999], and Han and Qiu [2007], among others. 4

6 agency-related changes in cash during each firm-year, where positive (negative) residuals indicate excessive (insufficient) changes in the cash balance. 5 Our paper is closely related to contemporaneous research by Oler and Picconi [OP 2010]. Several differences between our paper and OP emerge from our objectives to extend DRS and Xie [2001], which are not goals of OP. First, OP s model predicts optimal levels of cash and estimates the level of excess (insufficient) cash as the amount of a positive (negative) residual from the model. We predict the optimal change in cash and disaggregate the DRS change in cash variable into normal, excess and insufficient changes. Second, in the spirit of DRS and Xie [2001], our models predicting accounting and market performance examine differences in the persistence of and market efficiency with respect to the various components of current year earnings. Third, OP hypothesize that future accounting performance decreases with excessive or insufficient cash holdings. We hypothesize that the persistence of excess and insufficient cash components differ from each other and from the persistence of other components of free cash flow; i.e., the normal change in cash, distributions to debt holders and distributions to equity holders. Furthermore, we hypothesize that excess but not insufficient changes in cash potentially drive the DRS result indicating that the change in cash component of earnings has less persistence than other components of free cash flow. Fourth, we develop hypotheses based on predictions of differences in the persistence of the various earnings components that we examine. Finally, our paper includes the disaggregation of accruals, and we identify discretionary (or abnormal) accruals using the technique developed in Kothari et al. [2005]. Our paper then 5 Our paper differs from studies that evaluate earnings management through real activities manipulation and transaction timing (e.g., Roychowdhury [2006]; Cohen et al. [2008]; Gunny [2010]; Zang [2010]) and studies that evaluate cash flow management (e.g., Zhang [2006, 2009]; Hollie et al. [2010]). Both types of studies analyze behaviors that potentially affect the cash balance but, to our knowledge, no prior study evaluates the persistence and market pricing of changes in the cash balance that move the company towards (normal changes in cash) or away from (abnormal changes in cash) the optimal cash balance. 5

7 develops and tests predictions about the persistence and pricing of the disaggregated accrual and free cash flow components of earnings vis a vis one another. Our paper makes a significant contribution to the accounting literature, as, relative to DRS, we offer a more complete explanation for why changes in cash have less persistence than other components of free cash flow and earnings. We find that the hubris explanation offered by DRS only partially explains the lower persistence of the DRS change in cash variable, as both insufficient and excess changes in cash are associated with lower future earnings. Our paper also contributes hypotheses and tests of differences in persistence of the various components of the free cash flow and accrual portions of earnings, and we break up accruals into normal and abnormal (discretionary) components building on techniques developed by Xie (2001) and Kothari et al. [2005]. Our results may be summarized as follows. We replicate the DRS finding that the net change in cash earnings component has less persistence than the net distributions of free cash flow. We also replicate the results in Xie [2001] indicating that the abnormal (discretionary) accruals drive the evidence of lower persistence of accruals. In evaluating market efficiency with respect to the persistence characteristics of earnings components, we find evidence of market inefficiency with respect to the persistence of accruals and the mispricing of accruals is primarily driven by abnormal accruals. We find evidence that the market underreacts to the persistence of free cash flow. However, we fail to reject the null hypotheses that market rationally prices normal changes in cash and positive abnormal change in cash. The market s underreaction to the following three components of free cash flow drives the underreaction to the total: (i) underreaction to the persistence of negative abnormal changes in cash; (ii) underreaction to the persistence of net 6

8 distribution to debt holder cash flows; and (iii) underreaction to the persistence of net distribution to equity holder cash flows. The relatively large persistence of the distributions to investors and insufficient changes in cash components of free cash flow is consistent with our expectations. Insufficient changes in cash fail to provide for the firm s precautionary needs. As a result, the firm potentially misses profitable investment opportunities, and thus negative abnormal changes in cash correspond to lower future earnings. The market fails to anticipate this relation in pricing negative abnormal changes in cash. The market also apparently underreacts to the persistence characteristics associated with the good (bad) earnings prospects signaled by net distributions to (investments by) debt and equity holders. Our evidence does not support the DRS hubris hypothesis. We find no evidence of market inefficiency with respect to the persistence characteristics of excessively positive changes in cash, which might be used for investment in negative net present value projects. The rest of this paper is organized as follows. Regarding differences between the persistence and market pricing of various earnings components, Section 2 develops our expectations based on prior literature and our new hypotheses referring to normal and abnormal changes in cash. Section 3 describes the models used to: (a) estimate normal and abnormal accruals; (b) estimate normal and abnormal changes in cash; and (c) test our hypotheses. Section 4 describes our sample and data sources. Section 5 provides descriptive statistics. Section 6 presents and analyzes the results, and Section 7 concludes. 2. Hypotheses and Prior Literature 7

9 This section develops predictions about the relative persistence of the various earnings components. First, we evaluate the persistence of total accruals relative to the persistence of free cash flow, the two broadest components of earnings. As described by Richardson et al. [2006], accruals reflect three types of estimation error: first, accruals lack reliability due to inherent uncertainty about the future events subject to accrual accounting (e.g., unexpected warranty claims detracts from the persistence of the accrued warranty liability and corresponding expense); second, accruals lack persistence when managers use them to manipulate earnings, for example, towards bright line thresholds either through upward earnings management to meet the threshold or downward earnings management towards the threshold to build reserves; and third, accruals lack persistence if they reflect new investment and there are diminishing returns to investment. The persistence of free cash flow also suffers from manipulation by managers to achieve performance goals and diminishing returns to investment; however, free cash flows do not lack reliability due to any inherent uncertainty about the future affecting the actual amount of cash received or paid currently. Consistent with results in DRS and elsewhere, we expect that free cash flow has greater persistence than accruals. 6 Following DRS, we begin by decomposing free cash flow into changes in cash, distributions to debt holders, and distributions to equity holders. We then build on the model developed by BKS to further decompose the change in cash variable into normal and abnormal components, where abnormal changes refer to movements away from the optimum cash balance, and normal changes refers to precautionary movements toward the optimum cash balance. We extend Xie [2001] by disaggregating total accruals into normal and abnormal categories using techniques developed by Kothari, et al. [2005]. Normal accruals lack reliability due to the inherent 6 All hypotheses stated in the alternative form. 8

10 uncertainty related to estimates of the effects of future events, but do not lack reliability due to manipulation or diminishing returns to new investment. Therefore, we expect that, consistent with Xie s results, normal accruals have greater persistence than abnormal accruals. In accord with BKS, normal changes in cash move the company s cash balance towards an estimated optimal level. Positive (negative) normal changes result from increases (decreases) in the demand for cash holdings in the current period, as reflected in the changes in firms fundamentals. There is no clear a priori difference in the persistence of positive and negative normal changes in cash. On the other hand, we hypothesize differences in the persistence of positive versus negative abnormal changes in cash. Positive (negative) abnormal cash changes result in excessive (insufficient) cash balances. Compared to positive normal changes in cash which move the company towards the optimum, we expect positive abnormal changes to have lower persistence, since they represent movement towards excessive amounts of cash likely to be wasted on low NPV investments or unwarranted perquisites. Consequently, a dollar abnormal increase in cash is associated with a smaller increase in future earnings than a dollar normal increase in cash. Compared to negative normal changes in cash which move the company towards the optimum, we expect negative abnormal changes to be associated with even lower future earnings, because they move the company towards insufficient amounts of cash which could lead to missed opportunities to invest in positive NPV projects. Therefore, a dollar abnormal decrease in cash is associated with a larger decrease in future earnings than a dollar normal decrease in cash. Thus, we expect abnormal cash decreases to have greater persistence than abnormal cash increases. Figure 1 illustrates the hypothesized relations between future earnings, normal changes in cash, and abnormal changes in cash. Based on the above discussion, we propose the following hypothesis: 9

11 H1: Abnormal negative cash changes have greater persistence than abnormal positive cash changes. DRS find that changes in cash have almost identical persistence as accruals. They argue that changes in cash likely have low persistence because managers can waste cash on negative NPV projects, firms can window dress the balance sheet to improve perceived financial health; cash is also subject to manipulation; and cash increases can result in future expenditures on net operating assets that have diminishing returns to investment. However, DRS do not separate changes in cash and accruals into normal and abnormal components. The above arguments more likely apply to abnormal cash changes and abnormal accruals. Our next two hypotheses focus on the comparison of persistence of abnormal accruals and persistence of abnormal changes in cash. Assuming that DRS capture the average persistence level of changes in cash (i.e., normal changes, and positive and negative abnormal changes), we predict that positive abnormal changes in cash have less persistence than abnormal accruals because they are the least persistent component of changes in cash. Negative abnormal changes in cash, however, have greater persistence than abnormal accruals because they are the most persistent component of changes in cash. Our formal hypotheses are stated as follows: H2: Abnormal positive changes in cash have less persistence than abnormal accruals. H3: Abnormal negative changes in cash have greater persistence than abnormal accruals. 10

12 Finally, as described by DRS, firms with positive free cash flow have three choices: they can retain the cash, distribute the cash to equity holders, or distribute the cash to debt holders. Of the three options, DRS expect that retaining the cash has the least persistence because: (i) it may be wasted on negative net present value projects (Jensen [1986], Harford [1999]; (ii) it may represent a temporary increase due to earnings management through real activities, such as delaying R&D, advertising, maintenance, etc. (Roychowdary [2006], Gunny [2010]); (iii) it may be fraudulently misstated (e.g., Parmalat); or (iv) it may be a precursor to acquisition of assets with diminishing returns to investment. We expect that the above arguments apply relatively more to our measure of abnormal increases in cash. Therefore, we hypothesize that: H4: Abnormal increases in cash have lower persistence than normal changes in cash. Following the arguments in DRS, we also expect total changes in cash to have lower persistence than distributions to either debt or equity holders. Furthermore, consistent with the explanations in DRS, we expect net distributions to equity holders to occur only when the firm is relatively sure of sufficient future income to internally provide for the firm s cash needs and, therefore, we expect net distributions to equity holders to have greater persistence than net distributions to debt holders (also see Bartov [1991]). Also, as explained by DRS, these arguments are symmetric, since a firm with two choices for covering a free cash flow shortfall is more likely to issue equity than debt when the firm expects future losses. We also compare market efficiency with respect to the various earnings components. If the stock market tends to treat all components as having similar persistence characteristics, then we expect market underreaction to the pricing implications of higher persistence earnings 11

13 components, such as negative abnormal changes in cash and distributions to equity holders, and overreaction to the pricing implications of lower persistence components, such as abnormal accruals and positive abnormal changes in cash. More specifically, with reference to our contribution of the further disaggregation of the change in cash component of free cash flow, we hypothesize that: H5: Stock prices fail to fully impound the greater persistence characteristics of negative versus positive abnormal changes in cash. 3. Research Design Following DRS, we begin by decomposing companies income statements into accrual and free cash flow components as follows: 7 INCOME t = ACCRUALS t + FCF t, (1) where: the subscript refers to the fiscal year, INCOME t equals income before extraordinary items (IB), ACCRUALS represents total accruals and equals the change in non-cash (operating) assets (AT CHE) minus non-debt (operating) liabilities (LT DLTT DLC), and FCF represents free cash flow defined as INCOME t minus ACCRUALS t. 8 Next, we use the approach described in Kothari et al. [2005] to decompose total accruals into estimates of normal (i.e., non-discretionary) and abnormal (i.e., discretionary) components as follows. 7 The firm j subscript is suppressed in all models. Variable descriptions include parenthetical references to the relevant COMPUSTAT variable name. Unless otherwise specified, all variables are scaled by average total assets [(AT t + AT t-1 )/2]=NETASSETS t, where the subscript refers to firm j s fiscal year. 8 Throughout the paper cash refers to cash plus short-term investments (CHE), and non-cash assets refers to total assets less cash (AT CHE). This definition of cash proxies for all financial assets, and the definition of non-cash assets proxies for operating assets. 12

14 ACCRUALS t = α 0 + α 1 (1/ASSETS t ) + α 2 (ΔSALES t ΔAR t ) + α 3 PPE t + ε t (2) We cross-sectionally estimate model (2) for each industry (two-digit SIC code) and year requiring at least 10 observations in each industry-year. New variables in (2) are defined as follows: ΔSALES represents firm j s sales in year t minus sales in year t-1 (SALE t SALE t-1 ), ΔAR represents firm j s accounts receivable at the end of year t minus accounts receivable at the end of year t-1 (RECD t RECD t-1 ), and PPE represents firm j s gross property, plant and equipment at the end of year t (PPEGT t ). We subtract our initial estimate of firm j s normal (non-discretionary) accruals from model (2) from firm j s total accruals in year t in order to obtain an initial estimate of firm j s abnormal (discretionary) accruals for year t. We then subtract our initial estimate of firm j s abnormal accruals from the similarly derived estimate of abnormal accruals for firm k in year t, where firm k is the firm in firm j s industry with the closest year t return on total assets. This procedure provides a proxy for firm j s abnormal accruals, ABNACC jt, which we then subtract from firm j s total accruals, ACCRUALS jt to get our proxy for firm j s normal year t accruals, NACC jt. We disaggregate the free cash flow component in two steps. The first step follows DRS and the second step applies recent developments in the finance literature to further disaggregate the change in cash component of free cash flow into normal and abnormal components, where the normal (abnormal) component moves the firm s cash balance towards (away from) an estimate of the optimal level. The disaggregation begins with a reformulated balance sheet equation and the clean surplus relation. CASH + OPASSETS = OPLIAB + DEBT + EQUITY (3) 13

15 Equation (3) represents the left-hand-side of the balance sheet equation as cash plus all other assets, where cash includes the cash balance and all other financial assets and the rest of the firm s assets are labeled OPASSETS (i.e., total operating assets). The right-hand-side of the balance sheet equation includes operating liabilities (labeled OPLIAB), financial liabilities (labeled DEBT) and common stockholders equity (labeled EQUITY). 9 Subtracting operating liabilities from operating assets yields net operating assets (labeled NETOPASSETS), and representing the balance sheet equation as changes in each component creates a reformulated balance sheet equation in (4). CASH + NETOPASSETS = DEBT + EQUITY (4) Equation (4) represents the clean surplus relation; i.e., the change in common stockholders equity (labeled EQUITY) is completely explained by the difference between two components, INCOME and net distributions to common stockholders (labeled as INCOME DIST_EQ). EQUITY = INCOME DIST_EQ (5) Substituting the right-hand-side of (5) for EQUITY in (4), renaming the change in net operating assets ACCRUALS, renaming the change in debt DIST_D (i.e., net distributions to debt holders, and rearranging terms provides expressions for free cash flow on both sides of equation (6). 10 INCOME - ACCRUALS = CASH + DIST_D + DIST_EQ (6) Free cash flow created in the firm s operations equals the firm s income less accruals, and free cash flow available for distribution to providers of capital equals the change in cash plus net distributions to debt holders plus net distributions to common shareholders. DRS represent the 9 Preferred stockholders equity is included with DEBT. 10 Theoretically, interest expense should be removed from INCOME and considered part of DIST_D. Also, some portion of cash should be considered an operating asset and the other portion a financing asset. However, for practical purposes and following DRS, we treat interest expense and interest payable as operating items, and we treat all cash and other financial assets as financing items. For example, treating any portion of cash as an operating asset would require changes in that portion to be included in accruals and that would be inconsistent with the way we normally think of accruals. We do not expect these decisions to materially affect our inferences. 14

16 free cash flow variable in (1) as free cash flow available for distribution to all providers of capital, the right-hand-side of (6). We reproduce DRS INCOME disaggregation in (7) below. INCOME = ACCRUALS + CASH + DIST_D + DIST_EQ (7) In their study of whether the weights on factors hypothesized to affect optimal cash holdings have changed over time, BKS extend Opler et al. [1999] and develop a model of the optimal change in cash as follows. CASH t = α 0 + α 1 CASH t-1 + α 2 CASH t-1 + α 3 INDSIGMA t + α 4 MTB t + α 5 SIZE t + α 6 FCF t + α 7 NWC t + α 8 CAPEXP t + α 9 LEV t + α 10 R&D t + α 11 D DIV t + α 12 ACQEXP t + ε t (8) In (8) CASH t-1 and CASH t-1, together, proxy for the difference between the optimal and actual levels of the cash balance at the end of year t-1. The sum of CASH t-1 and CASH t-1, is highly correlated with the residual from a model predicting the optimal level of cash at the end of year t-1 and, therefore, we expect negative coefficients on CASH t-1 and CASH t-1 as firms holding excess (insufficient) cash at the end of year t-1 tend to move downward (upward) towards optimum levels during year t. INDSIGMA t represents the mean of the distribution of standard deviations of free cash flow, computed over the most recent five years (ending with year t-1), across all firms with firm j s two-digit SIC code. We expect a positive coefficient on INDSIGMA t as we expect firms in industries with more volatile free cash flows to hold more cash for precautionary reasons. MTB t represents the change in firm j s market to book ratio. We cannot predict the sign of the coefficient on MTB jt, because, depending on the firm s circumstances increases in the market-to-book ratio can proxy either for increased growth opportunities (Harris and 15

17 Marston [1994]) or for reductions in the firm s risk characteristics (Fama and French [1993], Fergusen and Shockley [2003]). In the former (latter) case, we would expect a positive (negative) coefficient on MTB jt as the need for precautionary amounts of cash increases (decreases) with growth opportunities (risk). SIZE t is measured as the natural log of the change in total assets. We expect a negative coefficient on SIZE t as larger firms benefit from economies of scale that reduce the need for cash. We expect a positive coefficient on FCF t as the change in cash is a positive component of this variable. NWC t represents the change in the firm s non-cash working capital. We expect a negative coefficient on NWC t as firms can substitute working capital for cash on hand. CAPEXP t represents the change in the firm s capital expenditures. This variable could have a positive coefficient on this variable as firms with growing capital expenditures keep more cash on hand as a precaution against failing to find financing for new capital expenditures. The variable could also have a negative coefficient as firms may draw on cash reserves to invest in long-term operating assets. LEV t represents changes in firm j s leverage, and we expect a positive sign as increased leverage increases the risk of failing to find cost effective financing for new investment opportunities. R&D t represents changes in R&D expenditures. This variable could have a positive sign as the firm takes precaution to have access to cash to finance growth opportunities created by investment in R&D. The variable might also have a negative sign as the firm draws on cash reserves to take advantage of potentially profitable R&D investments. 16

18 D DIV t indicates whether or not cash dividends on common stock increased during year t. This variable could have a positive coefficient on this variable as a firm increasing dividends needs to hold more cash in order to avoid missing a dividend payment. The variable might also have a negative coefficient as the firm draws on excess cash reserves to pay dividends. ACQEXP t represents the change in cash outflows on acquisitions, and this variable might have a positive coefficient as the firm needs more cash to avoid missing acquisition opportunities. On the other hand, the coefficient might be negative, if the firm draws on excess cash reserves to make acquisitions. Finally, the residual of the regression, ε t, represents the unexplained (abnormal) portion of the change in cash. We estimate (8) cross-sectionally for firms with the same two-digit SIC code, save the coefficient estimates and combine them with actual year t data to predict the change in cash in year t. We refer to the predicted change in cash as the normal change (N CASH), and we refer to the difference between the actual change and the predicted change as the abnormal change in cash holdings (ABN CASH). We now have proxies for all the variables in our fully disaggregated model of firm j s income for year t: INCOME = NACC + ABACC + N CASH + ABN CASH + DIST_D + DIST_EQ (9) To test our hypotheses about differences in the persistence of various combinations of the variables in (9), we estimate various forms of the following regression: + INCOME t+1 = β 0 + β 1 NACC t + β 2 ABACC t + β 3 N CASH t + β 4 ABN CASH t + β 5 ABN CASH - t + β 6 DIST_EQ t + β 7 DIST_D t + ε t+1 (10) 17

19 where ABN CASH + - t and ABN CASH t represent positive and negative ABN CASH, respectively. To evaluate market efficiency with respect to differences in the persistence of the various income components in (10), we adopt the Mishkin (1983) approach introduced to the accounting literature by Sloan (1996) and applied by DRS, among many others. We infer overweighting (underweighting) of a specific income component if the market attributes a higher (lower) valuation coefficient to it than the weight implied in its association with future income. We jointly estimate the following efficient forecasting and rational expectations pricing models: + INCOME t+1 = β 0 + β 1 NACC t + β 2 ABACC t + β 3 N CASH t + β 4 ABN CASH t + β 5 ABN CASH - t + β 6 DIST_EQ t + β 7 DIST_D t + ε t+1 (11) ARET t+1 = γ (INCOME t+1 - β * 0 - β * 1 NACC t - β * 2 ABNACC t - β * 3 N CASH t - β * + 4 ABN CASH t - β * 5 ABN CASH - t - β * 6 DIST_EQ t - β * 7 DIST_D t ) + ε t+1 (12) where ARET is annual buy-and-hold stock return calculated starting four months after the fiscal year-end, adjusted by the return on the CRSP size-decile portfolio in which the firm belongs. Market efficiency with respect to a specific component of income imposes the constraint that the valuation coefficient equals its counterpart in the forecasting model. This non-linear constraint requires that the stock market rationally anticipates the future income implications of each current income component. As in Mishkin (1983), we estimate (11) and (12) using iterative weighted nonlinear least squares. The test statistic is a likelihood ratio distributed asymptotically Chi-square (q): 2 n ln(ssr c / SSR u ) where: q = the number of constraints imposed by market efficiency 18

20 n = the number of observations in each equation SSR c = the sum of squared residuals from the constrained weighted system SSR u = the sum of squared residuals from the unconstrained weighted system We reject the rational pricing of any component of income if the above likelihood ratio statistic is sufficiently large. 4. Sample and data sources Examining our hypotheses regarding market efficiency with respect to persistence characteristics of earnings components requires access to corporate financial statement and returns data. Table 1 describes our sample selection process, which begins with all 398,649 firmyear observations on the 2008 COMPUSTAT annual database spanning the years We drop 94,074 observations in regulated industries (SIC codes and ). Relative to other industries, regulated industries, including the financial and utilities industries, have different earnings management incentives related to regulatory requirements and different financial statement structures. Next, we lose 202,556 (10,040) observations without sufficient data to estimate our model of normal changes in cash (accruals). Then we lose another 10,752 observations without one-year ahead earnings information. Finally, 26,630 observations do not have sufficient returns data to conduct our market efficiency tests, leaving us with a sample of 54,597 observations. (Insert Table 1 here) 5. Descriptive statistics and estimation of normal and abnormal changes in cash 19

21 Table 2 shows the results of estimating model (8) where we separate changes in cash into those that move the company towards the optimum, those that move the company s cash balance away from the optimum in the direction of excess cash, and those that move the company away from the optimum in the direction of insufficient cash. To estimate the optimal change in cash, we rely on the model developed in Bates et al. [2009] which, in turn, builds on the models of optimal levels of cash holdings developed in Opler et al. [1999]. (Insert Table 2 here) As expected, Table 2 shows that the coefficients on CASH t-1 and CASH t-1 are significantly negative, indicating that firms starting the year with either excess or insufficient cash tend to move towards a normal level during year t. 11 The coefficient on INDSIGMA is positive and significant, as expected, indicating that firms in industries with volatile cash flows tend to retain more cash for precautionary reasons. 12 The coefficient on MTB is positive and significant (at the 5% level, two tailed), indicating that the greater growth opportunities outweigh the lower risk for firms with higher market to book ratios in predicting the direction of the change in cash. As expected, the coefficient on SIZE is positive and significant, indicating that economies of scale allow larger firms to retain less cash. As expected, the coefficient on FCF is positive and significant, since the change in cash is a positive component of the change in free cash flow. As 11 Ideally, rather than CASH t-1 and CASH t-1, we would use the negative of the residual from a levels model of the normal cash on hand at the beginning of the year to indicate the direction of the year t change in cash for firms moving towards an estimated optimal level. However, using CASH t-1 and CASH t-1 to proxy for the direction and distance from the optimum at the end of year t-1 has the following advantages: (a) it follows the approach used in BKS; (b) it mitigates measurement error inherent in using the residual from a regression as a proxy for an underlying construct; and (c) CASH t-1 and CASH t-1 are highly correlated with the residual from a regression predicting the optimal level of cash at the end of year t Unless otherwise specified, significant means statistically significant at the 1% level (two-tailed). 20

22 expected, the coefficient on NWC is negative and significant, indicating that the relatively liquid nature of working capital creates a substitution effect between changes in working capital and changes in cash. The coefficient on CAPEXP is negative and significant, indicating that the need to draw on cash reserves outweighs any growth opportunities associated with large capital expenditures. Surprisingly, the coefficient on LEV is negative and significant, suggesting that countervailing forces outweigh the pressure to increase cash to avoid default as leverage increases. Finally, R&D expenditures, dividend payments and acquisition expenditures all use cash, so it is not surprising to find a negative relation between these three variables and the change in cash. Table 3 provides descriptive statistics for the whole sample and two subsamples. The first (second) subsample includes observations with positive (negative) residuals from model (8). While we see only a relatively small difference in total income between the two subsamples, Table 3, Panel D, shows that the two subsamples contain observations with quite different income component characteristics. In fact, the means and medians of all income components differ significantly between the two subsamples. Firms accumulating excess cash in year t [i.e., positive residual from model (8)] have significantly smaller accruals (both normal and abnormal) and significantly larger free cash flow. By construction, firms accumulating excess cash have larger total changes in cash, driven by the abnormal component. However, the predicted (i.e., normal) change in cash is smaller in the subsample with positive residuals from model (8). With respect to the rest of the components of free cash flow, the positive (negative) abnormal change in cash subsample has significantly larger distributions to debt (equity) holders. Finally, firms accumulating excess cash in year t have larger abnormal returns in year t+1. 21

23 (Insert Table 3 here) Table 4 shows a correlation matrix including all of the income component variables and the returns variable. On a univariate basis, relative to the relation between accruals and next year s earnings (0.12 Pearson correlation), we see a stronger relation between free cash flows and next year s income (0.41 correlation). Consistent with Sloan [1996], it looks like the market does not effectively distinguish differences in persistence of cash flows and accruals. The univariate results are consistent with market underreaction to the persistence of free cash flows and overreaction to the persistence of accruals, as we see a positive (negative) relation between free cash flow (accruals) and next year s returns. Consistent with Xie [2001], the abnormal component of accruals drives the negative relation between total accruals and future returns; whereas, consistent with DRS, net distributions to debt and equity holders drives the positive relation between free cash flow and future returns. Also, net distributions to equity holders drives the large persistence of free cash flow relative to other earnings components. (Insert Table 4 here) Consistent with the descriptive statistics in Table 3, when we split the samples based on the sign of abnormal changes in cash we see large differences in univariate correlations between the variables. In particular, consistent with our arguments in support of H3 in the positive (negative) residual subsample, the abnormal change in cash is negatively (positively) associated with next year s income. Furthermore, it appears that the market does not recognize these differences in persistence, as Table 4 Panel B (C) shows that the abnormal change in cash is negatively 22

24 (positively) correlated with future returns. The univariate results should be read with caution, since Table 4 shows the well-known negative relation between free cash flow and accruals ( correlation). 6. Results 6.1 Persistence results Table 5 reports the first set of tests of our hypotheses. As described in Table 5, INCOME is highly persistent, with a coefficient of 0.741, which approaches the equivalent of a random walk model of annual earnings (e.g., Lookabill [1976]). However, as hypothesized, the persistence of the components of earnings differ from each other. Consistent with DRS, Sloan [1996] and others, relative to accruals, Table 5 shows a significantly higher persistence coefficient on free cash flow (0.752 versus 0.672). Similarly, consistent with Xie [2001] and DRS, respectively, normal levels of accruals have higher persistence than abnormal accruals (0.558 versus 0.536), and the persistence of distributions to equity holders (0.711) drives the higher persistence of free cash flow. (Insert Table 5 here) With respect to our new variables, normal and abnormal changes in cash, consistent with H1, we find significantly higher persistence of negative abnormal changes in cash than positive abnormal changes in cash (0.588 versus 0.421). We attribute this finding to the hypothesized general condition that, as the firm moves away from the estimated optimal amount of cash, predictions of future earnings decline. That is, as the firm accumulates excess cash, it becomes 23

25 more likely that managers will invest in low NPV projects or consume the cash in perquisites. Similarly, as the firm accumulates less cash than the predicted optimum, managers may have to forego investment in positive NPV projects. Consistent with H2 and H4, respectively, we find that positive abnormal cash changes have significantly less persistence than abnormal accruals (0.421 versus 0.538) and normal changes in cash (0.421 versus 0.556). Consistent with H3, we find stronger persistence of negative abnormal cash changes than abnormal accruals (0.588 versus 0.538, significant at the 5% level, one tailed). Finally, consistent with the arguments and evidence in DRS, changes in cash have significantly less persistence (0.610) than the persistence of net distributions to debt holders (0.699) which in turn have significantly less persistence than net distributions to equity holders (0.744). Overall, our results are consistent with prior research, and the variables we introduce to literature related to the persistence of earnings components behave according to our hypotheses. Our new variables provide more detailed analysis of the change in cash component of free cash flow. Essentially, we use the BKS model of optimal changes in cash to split each firm-year s change in cash into three components: normal changes in cash (the model s predicted amount), abnormal increases (positive residuals from the model) and abnormal decreases (negative residuals from the model). Relative to abnormal accruals, normal changes in cash and abnormal cash decreases, abnormal increases in cash have low persistence in relation to next year s earnings, presumably due to costs associated with moving away from the estimated optimum and towards excessive cash balances. Relative to normal and insufficient changes in cash, excessive cash increases are likely to be invested in low or perhaps even negative return projects and to provide unneeded perquisites for the firm s employees/managers. Thus, excessive increases in cash have low persistence as they are associated with relatively low future income. On the other 24

26 hand, insufficient cash increases (negative residuals from the model) have relatively high persistence as they, too, lead to lower future income due to costs associated with missed investment opportunities. 6.2 Market pricing results Table 6 shows the results of our tests of the market efficiency hypothesis (H5) using the Mishkin test introduced by Sloan [1996] and used in Xie [2001], DRS and many other related papers. Note that we report mean annual coefficients in Table 5; whereas, consistent with prior research, Table 6 reports Mishkin test results using the entire sample pooled across firms and over time. Therefore, the actual persistence parameters reported in Table 6 are somewhat different from those in Table 5. (Insert Table 6 here) Panel A of Table 6 presents results for the basic income autoregression model. The persistence coefficient on INCOME is If the market correctly anticipates the persistence of current income, then the implied persistence parameter should be the same as the actual persistence parameter. However, we find that the implied persistence parameter is 0.697, which is significantly less than It appears that the market underreacts to the overall persistence of income. This result contrasts with Sloan s [1996] and DRS s findings, possibly due to differences in sample composition. Panel B presents the results for the test of market efficiency with respect to accruals and free cash flow components of income. Consistent with the results in Table 5, accruals are 25

27 significantly less persistent than free cash flows (0.684 versus 0.774). We find that the implied persistence coefficient of accruals in stock prices is significantly higher than the actual persistence coefficient (0.750 versus 0.684); whereas the implied persistence coefficient of free cash flows in stock prices is significantly lower than the actual persistence coefficient (0.614 versus 0.774). These results are consistent with Sloan [1996], suggesting that the market overestimates the persistence of accruals but underestimates the persistence of free cash flows. Following DRS, Panel C further decomposes free cash flows into changes in cash, net distributions to equity holder and net distributions to debt holder. We continue to find that net distributions to equity holders have the highest persistence level among all components of earnings (0.789), followed by net distributions to debt holders (0.671). The corresponding implied persistence parameters in stock prices are for net distributions to equity holders and for net distributions to debt holders. It appears that the market underreacts to the persistence of these two free cash flow components. However, we find that the implied persistence coefficient of changes in cash is not significantly different from the actual persistence coefficient of changes in cash (0.557 versus 0.573), suggesting that the market anticipates the lower persistence of changes in cash and efficiently incorporates the information in stock price. Panel D reports the results after we disaggregate accruals and changes in cash into normal and abnormal components. Consistent with Xie [2001], we observe market overpricing of abnormal accruals, but not normal accruals. We also fail to reject the hypothesis that the market efficiently anticipates and prices the persistence of normal changes in cash. However, the market seems to be significantly less efficient in pricing the abnormal changes in cash (0.344 implied market persistence coefficient versus actual persistence coefficient). 26