A New Approach to Empirical Analysis of the Relation between Change in Value and Earnings

Transcription

1 A New Approach to Empirical Analysis of the Relation between Change in Value and Earnings Peter Easton Center for Accounting Research and Education University of Notre Dame Peter Vassallo Australian School of Management University of New South Wales Eric Weisbrod School of Business Administration University of Miami September 2015 We thank Bill Beaver, Demetris Christodoulou, Mincherng Deng, Joseph Gerakos, Aloke Ghosh, Adel Ibrahim, Chris Jones, Sok-Hyon Kang, Yaniv Konchitchki, Bin Li, Steve Lilien, David Lont, Martein Lubberink, Steve Monahan, Jim Ohlson, Suresh Radhakrishnan, Sundaresh Ramnath, Oded Rosenbaum, Greg Sommers, Theodore Sougiannis, Michael Turner, Oktay Urcan, Terry Walter, Bill Wright, Nir Yehuda, Xiao-Jun Zhang, and seminar participants at Baruch College, George Washington University, INSEAD Accounting Symposium, Singapore, Southern Methodist University, the University of Adelaide, the University of California at Berkeley, the University of Illinois at Urbana-Champaign, the University of Miami, the University of Otago, the University of Queensland, the University of Sydney, the University of Texas at Dallas, and the Victoria University of Wellington for helpful comments on an earlier draft.

2 A New Approach to Empirical Analysis of the Relation between Change in Value and Earnings Abstract We suggest and show the efficacy of two fundamental changes to the methodology at the core of the vast empirical literature examining the extent to which accounting captures concurrent changes in market value. First, we focus on the part of the earnings/returns relation that is not dollar-for-dollar because, at best, the part that is recorded dollar-for-dollar is uninteresting empirically and, at worst, including this part may lead to incorrect inferences. Second, we suggest the inclusion of an omitted variable, capturing transactions with owners, in the earnings/change in value relation. Absent unconditional conservatism, this additional variable should provide no explanatory power in the model we propose. However, our empirical analyses demonstrate that this added variable contributes similar explanatory power for earnings to that provided by cum-distribution value change, and that the sign of this variable also affects estimates of conditional conservatism derived from either our model or that of Basu [1997]. Keywords: Accounting Conservatism, Firm Growth, Free Cash Flows, Earnings Return Relation

3 1. Introduction We suggest two fundamental extensions to the methodology that is the basis of a vast empirical literature where the question at issue is the extent to which accounting captures concurrent changes in market value. This literature, particularly since Beaver, Lambert and Ryan [1989] and more recently, Basu [1997] is generally based on a regression of earnings on returns, with a dummy intercept and slope variable to identify bad news (negative returns) vs. good news (positive returns); the estimates of the earnings/returns coefficients capture the portion of the returns that are captured in earnings of the fiscal period. We introduce two refinements. First, we focus on the part of the earnings/returns relation that is not, as a reasonable first approximation, dollar-for-dollar because, at best, the part that is recorded dollar-for-dollar is uninteresting empirically and, at worst, including this part may lead to incorrect inferences. 1 Accordingly, we examine the relation between enterprise profit after tax (EPAT, often referred to as operating income), and changes in enterprise value. 2 Second, we add financial flows into and out of the enterprise (enterprise cash flows, or ECF) as an additional explanatory variable, based on the observation that changes in enterprise market value contain two components; value generated by enterprise assets in place, and flows of value into and out of the enterprise. Extant literature focuses on the relation between earnings and generated changes 1 This portion is uninteresting empirically because we know that the accounting (e.g., the effective interest method) is such that the mapping from change in value to earnings is dollar to dollar and there is nothing to be learned from estimating a known relation. 2 We use the word enterprise to identify the entity owned by the equity and debt holders, which engages in the firm s primary revenue-generating activities. We follow the terminology of Gode and Ohlson [2013] to distinguish between enterprise activities and financing activities. Enterprise activities are sometimes referred to as operating activities in practice, and EPAT is sometimes referred to as net operating profit after tax (NOPAT) (see, for example, Penman [2012]). See Appendix A.1: Terminology for a quick reference of common terms employed in the paper. Enterprise/operating activities include activities such as research, development, purchasing inputs, production, marketing, and distribution of products and services (e.g., coffee in the case of Starbucks, passenger miles in the case of United Airlines, and brand-name household consumer products in the case of Procter and Gamble). 1

4 in equity value (i.e. cum-dividend stock returns), and assumes that this relation is unaffected by transactions with stockholders during the period. We demonstrate that accounting principles, specifically aspects of accounting often labelled as unconditional conservatism (e.g. Ryan [2006]), motivate the inclusion of the omitted variable, ECF, in tests designed to examine the extent to which accounting captures concurrent changes in enterprise market value. In fact, we show that enterprise cash flow contributes similar explanatory power for earnings to that provided by enterprise value generated or lost during the period. To elaborate, the summary income statement variable that is pervasively studied in marketbased accounting research is earnings available to common shareholders. These earnings are comprised of a portion, EPAT, which generally records value change at less than dollar-fordollar and a remaining portion, net financial expenses (NFE), which, as a reasonable first approximation, records value change dollar-for-dollar. The right-hand side of the earnings/return relation is cum-dividend returns to common equity shareholders. These returns are a function of return on the enterprise and return on net financial assets or liabilities. Returns on financial assets and liabilities are, as a reasonable first approximation, recorded in the financial statements dollar for dollar, while the enterprise portion is generally recorded at less than dollar-for-dollar (i.e., conservatively). It follows that studies examining the extent to which earnings capture value change in a timely fashion should remove the components of earnings that record change in value dollar-for-dollar from the analysis. This motivates our focus on the relation between EPAT and change in enterprise value ( EV). Thus, the focus of our paper is on the recording of change in the market value of a corporate enterprise in the financial statements of the corporation. From the perspective of the owners of the enterprise, EPAT is the key accounting measure of cum-ecf value generated by the 2

5 enterprise during a period. 3 But, the accounting measure, EPAT, captures only a portion of the cum-ecf market value of the enterprise generated during the period, which we label enterprise value generated (EVg). 4 A key point of our paper is that EPAT, as well as the relation between EPAT and EVg, varies with the degree to which overall change in the market value of the enterprise during the period ( EV) is generated from assets already in the enterprise (EVg) or occurs via investments by (or distributions to) the owners of the enterprise (ECF). Our paper describes and empirically identifies these differences in accounting across the two sources of change in enterprise value. We argue and show that accounting for change in value depends on both the direction (expansion vs. contraction) and the source of the overall enterprise value change. Further, we demonstrate that this is because, in our corrected model, the relation between EPAT and EVg is largely determined by accounting principles that fall under the oftenused label conditional conservatism, while the relation between EPAT and ECF isolates the effects of accounting principles more generally labelled unconditional conservatism. 5 Our main contributions are to refine the analysis of conditionally conservative accounting, and to capture the effects of unconditional conservatism in determining the extent to which accounting captures concurrent changes in market value; importantly, we argue and show that conservative accounting for EVg differs from the conservative accounting for ECF. 3 EPAT is the enterprise-level equivalent of earnings available to common shareholders, which, in turn, is the key accounting measure of cum-dividend change in the value of shareholders equity during the period. 4 This variable is the enterprise-level equivalent of cum-dividend (i.e. buy-and-hold) equity returns, and follows from the market identity EVg ΔEV + ECF. 5 Our inclusion of enterprise cash flows in our analyses further motivates our examination of the relation between accounting and changes in market value at the enterprise rather than the equity level, which may seem to some as an unnecessarily large departure from prior literature. Enterprise cash flows are more appropriate because cash flows to/from equity holders often have nothing to do with growth in the value of the enterprise and, in such cases, may have no effect on earnings at all (e.g., debt/equity swaps, increasing debt to pay dividends, etc). 3

6 Our empirical analyses are based on a regression of the accountants measure of change in enterprise value (EPAT) on generated change in the market value of enterprise assets (EVg) and net contributions from/distributions to the owners of the enterprise (ECF), with each of these variables deflated by beginning-of-year enterprise value (EV). This regression specification permits a study of conditional conservatism (via the estimate of the coefficient on EVg) and unconditional conservatism (via the estimate of the coefficient on ECF) within the same regression. We show that estimates of the coefficients relating EPAT to EVg and EPAT to ECF vary in predictable ways according to the signs of ΔEV (i.e., increase vs. decrease in enterprise value), ECF (i.e., value provided by vs. value distributed to the owners of the enterprise during the period) and EVg (i.e. the enterprise-level equivalent of the Basu [1997] partition). We show that enterprise cash flow, ECF, provides significant incremental explanatory power for EPAT beyond EVg, and that the mapping from EVg to EPAT depends on the sign of ECF. Our results show that, in most cases, the association between ECF and EPAT is as large as, and sometimes larger than, the association between EVg and EPAT. We find that, on average, a one standard deviation change in ECF (EVg) is associated with approximately a 0.30 (0.18) standard deviation change in EPAT. In other words, we show that accounting measures of enterprise growth are often more strongly associated with transactions between the enterprise and its owners during the period than with value generated by the enterprise during the period. 6 We compare our empirical analyses to those in the extant literature in order to illustrate the effects of our suggested methodological refinements of: focusing on the accounting at enterprise 6 Our analysis of ECF departs from the extant literature that considers the mapping from changes in value to cash flows rather than the mapping from cash flows to earnings (e.g., Collins et al. [2014], who examine the mapping from equity returns to operating cash flows, and Penman and Yehuda [2004], who examine the mapping from enterprise cash flows to change in enterprise value). As far as we are aware, this mapping, which explains much of the cross-sectional variation in EPAT, has not yet been considered in the literature on accounting for value change. 4

7 level rather than at the equity level; and, the addition of free cash flows to/from the enterprise as an additional explanatory variable. We provide examples of partitions of the data where these refinements are predicted and shown to be more important. The extant literature provides evidence that conditional conservatism estimated via the regression of earnings on returns varies with market capitalization, balance sheet conservatism (i.e., the ratio of the book value of equity to market value of equity) and level of debt (see Kahn and Watts [2009], who identify these three attributes as core indicators of conditional conservatism). We demonstrate the importance of these characteristics in our new specification. Most, but not all, of the results that are observed at the equity level (via analyses of the estimates of the earnings/return coefficients) are also seen when we focus on the EPAT/EVg coefficient. We also demonstrate the added understanding of the role of size, book-to-market and debt that may be gained by addition of ECF to the analysis. In order to demonstrate the methodological gains from incorporating unconditional conservatism into the analyses, we also show how the estimate of the coefficient relating EPAT to ECF and (of lesser importance) the coefficient relating EPAT to EVg changes with intangibles intensity, since intangibles are assets that are accounted for unconditionally conservatively. The remainder of the paper proceeds as follows. We begin, in section 2, by describing our empirical method and predictions. In doing so, we demonstrate our reasons for our focus on accounting at the enterprise level and for the addition of ECF. We describe our sample selection procedure in section 3, and discuss selected descriptive statistics. Section 4 analyzes and compares partitions of the data based on whether enterprise value is increasing or decreasing and on the source of this value increase/decrease. In Section 5, we examine whether the empirical 5

8 manifestation of change in enterprise value in the financial statements differs as expected across firm-year characteristics. Section 6 presents a summary and conclusions. 2. Empirical Method and Predictions In this section we first show how we can obtain more focused analyses of conservative accounting by undertaking the analyses at the enterprise level rather than at the equity level. Then we provide reasons why enterprise cash flows should be included as an additional variable in the analyses Changing the Focus from Equity Level Analyses to Enterprise Level Analyses The analyses in Basu [1997] and the papers that follow are based on a regression of earnings on cum-dividend returns with a dummy and intercept that is one if returns are negative, zero otherwise. The relation between earnings and returns, without the dummy terms, may be written: 7 it = α 0t + α 1t ε 1it (1) where it is comprehensive net income available to common shareholders of company i for year t, it is the market value of equity in company i at the end of year t, and it is free cash flow to/from equity holders of company i in year t. it is the net of all transactions with 7 Basu [1997] and papers that follow alternately use both comprehensive net income deflated by market capitalization (as we do in equation (1)) and earnings before extraordinary items per share deflated by beginning of year price per share. All of these studies, as far as we are aware, use compounded monthly cum-dividend stock returns as the independent variable. We begin this section with the variables in equation (1) because it makes our algebra more succinct. In our empirical analyses, when making comparisons with existing methodology, we use scaled earnings before extraordinary items and compounded CRSP stock returns in our analyses (e.g. Ball, Kothari, and Nikolaev [2013]). Inferences remain unchanged if we substitute Compustat-based measures of comprehensive net income, fiscal-year change in market value of equity, and free cash flows to equity holders. 6

9 equity holders (cash dividends plus payments for stock repurchases minus proceeds from new stock issues). Recognizing the facts that: (1) net income is equal to enterprise profits after tax less net financial expenses; (2) the market value of equity is equal to the market value of the enterprise minus the market value of net financial liabilities (net debt) 8 ; and (3) free cash flow to equity holders is equal to enterprise cash flows (ECF) minus the portion of ECF that goes to/from holders of net debt, equation (1) may be re-written: it it = α 0t α 1t ε 1it (2) where it is the enterprise profit after tax of firm i for year t, it is net financial expenses of company i for year t, it is the market value of the enterprise (operations) of firm i at time t, it is the market value of net debt of company i at time t, it is enterprise cash flows to the owners of the enterprise (equity and net debt holders), and it is cash flows to/from the net debt holders. As a reasonable first approximation, net financial expenses are accounted for dollar for dollar (in other words there is no conservatism); that is, NFE = ΔMVD + FCFD. It follows that α 1t, the coefficient of interest in extant literature, captures a combination of mappings from change in value to earnings that are, as a first approximation not conservative (NFE = ΔMVD + FCFD) 8 Note that the market value of net debt (as well as net financial expenses) can be negative. That is, the equity holders can own the enterprise and a pile of cash (as in Microsoft, for example). It follows that substituting MVE with EV - MVD (where negative MVD implies market value of net financial assets) remains valid. 7

10 along with the mappings that are conservative (EPAT ΔEV + ECF). So, instead, we base our analyses on the following relation: it = β 0t +β 1t ε 3it (3) We call the sum of ΔEV and ECF enterprise value generated, EVg. Note that EVg is the cum-ecf return on enterprise value. In other words, EVg is the enterprise-level equivalent of cum-dividend equity return. Similarly, EPAT is the enterprise-level equivalent of comprehensive net income available to common shareholders. 2.2 Estimating Conditional Conservatism at the Enterprise Level Compare the estimates of the coefficients from the following equity level analyses: it = γ 0t + γ 1t γ 2t γ 3t ε 4it (4) where = 1 if is negative, 0 otherwise, with the estimates from the enterprise level analyses: it = δ 0t + δ 1t +δ 2t δ 3t ε 5it (5) where = 1 if is negative, 0 otherwise. We show how inferences based on estimates of δ 3t may differ from inferences based on estimates of γ 3t. The essence of the argument in Basu [1997] is that conditional accounting conservatism leads to more of bad news (negative returns) than good news (positive returns) being captured in earnings of the return period; in particular differences in γ 3t across samples of observations (e.g., across accounting regimes) capture differences in conditional conservatism. This argument also 8

11 applies at the enterprise level. But, our refinement, at the enterprise level, focuses the analyses on the entity where conditional conservatism will be manifested. The estimate of the coefficients on EVg (i.e., δ 2t and δ 3t ) capture conditional conservatism in the accounting for enterprise value generated from assets in place, which is calculated as total change in enterprise value (i.e., ΔEV) before the effects of any zero-npv changes in enterprise value due to contributions by or distributions to the owners of the enterprise, ECF. By adding back the nominal value of ECF to ΔEV, EVg also captures conditional conservatism due to the investment of ECF in non-zero NPV projects in current or prior periods. We first illustrate conditional conservatism with respect to enterprise assets in place and then discuss how EVg also captures conditional conservatism in the accounting for ECF. 9 Increase in enterprise value of assets in place (i.e., positive EVg) due to such things as discovery of new technology, acquisition of new contracts, which may be serviced with the existing enterprise assets, effective cost-cutting, re-investment of internally generated cash in positive NPV projects, etc., may affect both current EPAT and future EPAT. The change in enterprise value will reflect the present value of the effect on current and future EPAT. The 9 Some examples of generated value change include increases in expected asset turnover or profit margin, new product developments, increased brand recognition, and increased growth options due to weakening of competition or increases in the size of the economy as a whole. Stated differently, we define generated value change as any increase in the market value of the enterprise beyond the value change (growth) due to cash inflows from owners during the period. Similarly, a generated decrease in enterprise value could be driven by the same economic factors cited as growth examples. Notably, our definition of generated growth includes growth from what Roychowdhury and Watts [2007] refer to as rents and Ball, Kothari and Nikolaev [2013] label as the g component of stock returns (which is associated with revisions in the value of growth options or un-booked intangibles ), as well as curtailment, which is the focus of Lawrence et al. [2015]. Generated value change may also be associated with the investment of ECF (i.e., contributed growth) in non-zero NPV projects, modelled in Feltham and Ohlson [1995] and [2000]. 9

12 relation between current EPAT and positive EVg (i.e., δ 2t ) will reflect the portion of the change in value that is captured in current EPAT. Decrease in enterprise value of assets in place (i.e., negative EVg ) due to such things as loss of comparative technological advantage, loss of market share, cost increases, etc., may affect both current EPAT and future EPAT. Since these effects are more likely to be transitory rather than permanent on average (otherwise the firm will go out of business), the mapping from negative EVg to EPAT (i.e., δ 2t + δ 3t ) is likely to be greater than the mapping from positive EVg to EPAT (i.e., δ 2t ) Further, generally accepted accounting principles place greater verification thresholds on increases in value recorded in EPAT than on decreases in value recorded in EPAT, which can also contribute to a greater portion of EVg being recognized in current-period EPAT when EVg is negative. 10 Note that EVg also captures conditional conservatism with respect to value generated (or lost) from ECF if the ECF is invested in a non-zero NPV project. By adjusting ΔEV for the nominal amount of ECF, dollar-for-dollar changes in enterprise value from zero-npv investments (or divestments) of enterprise assets during the period are removed from EVg. However, EVg will still contain the non-zero-npv nominal effect of investments (or divestments) of enterprise assets during the period. The portion of these deviations from zero- NPV that are captured in the current EPAT will depend on whether the NPV is positive or negative. Positive NPV is more likely to be due to changes in future profitability many years into the future, and, hence, the estimate of the coefficient δ 2t will be smaller whereas a negative 10 In other words, negative EVg may be recognized for accounting purposes before the decreases in expected future enterprise value are realized (e.g. asset write-downs), while increases in expected enterprise value are not recognized for accounting purposes until the expected outcomes are realized. 10

13 NPV investment will likely effect EPAT of the current period (e.g., there will be a write-down) and hence the mapping from negative EVg to EPAT (i.e., δ 2t + δ 3t ) when there is negative NPV investment/divestment of ECF will be higher. Thus, the coefficients δ 2t and δ 3t also capture conditional conservatism due to non-zero-npv investment or divestment of ECF during the period. Again, this asymmetric accounting treatment is in the spirit of Basu [1997]. 2.2 Conservatism in the Accounting Treatment of Enterprise Value Generated differs from the Conservatism in the Accounting Treatment of Enterprise Cash Flows It is well known that, even in the absence of positive or negative NPV projects, and/or changes in expectations about the value of enterprise projects, accounting can still be conservatively biased due to the mechanical application of accounting rules. This is often referred to as unconditional accounting conservatism (e.g. Ryan [2006]), and, on average, leads to a downward bias in EPAT when the enterprise is growing due to injection of ECF and an upward bias during periods of enterprise contraction due to ECF outflows. 11 In the presence of unconditional conservatism, ECF represents an omitted variable in the EPAT/EVg relation. In order to capture the effect of unconditional conservatism, we add ECF to equation (3) as an additional explanatory variable. The mapping from ECF to EPAT in the revised model reflects unconditional conservatism, which we define as accounting conservatism that leads to under-statement of EPAT (overstatement of expenses) and this understatement is independent of whether value is generated vs. lost. We consider two examples of unconditional conservatism; 11 See, for example, textbook materials such as Penman [2012] or Gode and Ohlson [2013], as well as studies such as Feltham and Ohlson [1995], Easton and Pae [2004], Easton [2009], and Ohlson [2009]. 11

14 where ECF is negative (the owners of the enterprise are investing in the enterprise) and where ECF is positive (the owners of the enterprise are taking cash out of the enterprise). If there is ECF inflow, unconditional conservatism will be reflected in the amount of the ECF that is expensed (rather than capitalized). Investment of ECF will increase in the market value of the enterprise, ΔEV, by the amount of the ECF inflow (i.e. negative ECF). And, if there is no conservatism in accounting (i.e., ΔNEA = ΔEV, where ΔNEA is the change in the book value of net enterprise assets), ΔNEA will also increase by the amount of negative ECF, and the accounting identity, EPAT ΔNEA + ECF ensures that ECF has no effect on EPAT. 12 But, conservative accounting will lead to greater expensing during the period (e.g., complete expensing during the period with R&D, or partially accelerated expensing in any case in which unconditionally applied accounting depreciation is greater than depreciation in market value of the asset) and ECF will reduce EPAT of the period (but not EVg). The effect of unconditional accounting conservatism is similar if there is ECF outflow funded by, say, sale of enterprise assets. If the asset sale price is equal to the book value of the asset (i.e., there is no conservatism in the accounting for the value of the asset), ΔNEA = ΔEV = ECF, ECF has no effect on EPAT. But, if the book value is less than the salvage (sale) value there will be an accounting gain on sale, which will increase EPAT but not EVg. Note that unconditional accounting conservatism reduces EPAT relative to EVg when ECF is negative and increases EPAT relative to EVg when ECF is positive. 12 The market identity EVg ΔEV + ECF ensures that zero-npv ECF will not affect EVg, regardless of the accounting treatment of ECF. 12

15 2.3 Putting it all together Accordingly, we analyze the following model across a number of partitions based on the signs of EVg, ECF, and ΔEV: it = θ 0t +θ 1t +θ 2t ε 6it (6) 13 For expositional convenience in the remainder of the paper, we will refer to the variables it, it and it as EPAT, EVg and ECF when discussing empirical results. Formal variable definitions for all variables used in our empirical analyses, including the data items used to compute each variable, also appear in appendix A.2. In regression (6), the estimate of the coefficient θ 1t (across partitions) captures conditional conservatism while the estimate of the coefficient θ 2t captures unconditional conservatism. To see this, recall the accounting identity EPAT ΔNEA + ECF and market identity EVg ΔEV + ECF. When there is no conservatism in accounting (i.e., ΔNEA = ΔEV), these identities converge and EPAT is equivalent to EVg, resulting in a coefficient of 1 on EVg and 0 on ECF in equation (6). As described in section 2.1, conditional conservatism leads to a coefficient θ 1t less than 1, but which is expected to be larger 13 Since EPAT ΔNEA + ECF this regression may be re-written: it it = θ 0t +θ 1t +θ 2t ε 6it (7) Alternatively, it = θ 0t +θ 1t + 1 θ 2t ε 6it (8) Regression (8) permits a focus on change in the value of enterprise assets on the balance sheet and on the portion of the variable EPAT where there is conservatism. Despite this advantage, we do not run this regression for two reasons: (1) the coefficients from regression (8) may be readily calculated from those in regression (6) and at times we do these calculations; and, (2) we choose to depart from the extant literature by replacing equity earnings with enterprise earnings rather than by the change in consecutive balance sheet amounts. 13

16 for negative changes in expected value than positive changes in expected value. Because deviations from zero-npv ECF are captured in EVg, only unconditional conservatism, the mechanical portion of accounting conservatism affecting ECF flows into and out of the enterprise independent of EVg will cause θ 2t to differ from zero. As described in section 2.2., the accounting equation dictates that unconditional conservatism creates a positive association between EPAT and ECF. This leads to a coefficient on θ 2t greater than zero, which is likely to vary with the sign of ECF. Our analysis considers interactions between conditional and unconditional conservatism. Accordingly, we examine six partitions formed on the sign of each independent variable in equation (6) as well as the sign of combined change in enterprise value from both sources (ΔEV), resulting in a total of six sub-samples: (1) Enterprise Growth, Value Creation, Net Cash Inflow (+ΔEV, +EVg, ECF in): there is growth on every dimension (i.e., all is going well and there is a further injection of cash); (2) Enterprise Growth, Value Creation, Net Cash Outflow (+ΔEV, +EVg, ECF out): the enterprise is growing but there is net cash outflow (i.e., all is going well and there is a removal of cash); (3) Enterprise Growth, Value Loss, Net Cash Inflow (+ΔEV, -EVg, ECF in): the enterprise is growing because of net cash inflow despite negative internal growth (i.e., the enterprise growth is due to the injection of cash by its owners): (4) Enterprise Contraction, Value Creation, Net Cash Outflow (-ΔEV, +EVg, ECF out): the enterprise is contracting because net cash outflow is greater than value generated (i.e., 14

17 generated growth is not large enough to replace the value that the owners are taking out of the enterprise); (5) Enterprise Contraction, Value Loss, Net Cash Inflow (-ΔEV, -EVg, ECF in): the enterprise is contracting despite net cash inflow (i.e., the enterprise is fairing badly and there is an injection of cash) and, (6) Enterprise Contraction, Value Loss, Net Cash Outflow (-ΔEV, -EVg, ECF out): there is contraction on every dimension (i.e., the enterprise is fairing badly and there is removal of cash). 2.4 Conservative Accounting and Enterprise Characteristics The extant literature provides evidence that conditional conservatism estimated via the regression of earnings on returns varies with market capitalization, balance sheet conservatism (i.e., the ratio of the book value of equity to market value of equity) and level of debt (see Kahn and Watts [2009], who identify these three attributes as core indicators of conditional conservatism). We demonstrate the importance of these characteristics in our new specification by showing how the estimates of the coefficients on EVg and ECF vary with these enterprise characteristics. We also demonstrate the added understanding of the role of size, book-to-market and debt that may be gained by addition of ECF to the analysis. In order to demonstrate the methodological gains from incorporating unconditional conservatism into the analyses, we also show how the estimate of the coefficient relating EPAT to ECF and (of lesser importance) the coefficient relating EPAT to ΔEV changes with intangibles intensity, since intangibles are assets that are accounted for unconditionally conservatively. 15

18 3. Sample Selection and Selected Descriptive Statistics We obtain annual financial statement data from the Compustat (Xpressfeed) database for fiscal years We match this with stock price return data from the CRSP database. The sample period begins in 1963 in order to ensure data availability in Compustat. We exclude foreign incorporated firms (we require that Compustat FIC=USA), financial institutions (SIC codes ), utilities (SIC codes ), observations with negative market value or total assets (potential data errors), and observations with beginning-of-fiscal-year stock prices less than one dollar. Following the method in several textbooks on financial statement analysis and valuation, net contributions, ECF, are calculated from income statement and balance sheet data as enterprise profit after tax, EPAT, minus the change in net enterprise assets, ΔNEA. 14 We require that all observations included in the sample have sufficient data available for the calculation of all variables in Table 1. To mitigate the influence of extreme observations on our results, we truncate observations that fall in the top or bottom 1 percent of any of the variables included in the primary regression equation (equation 6) as well as price-deflated earnings and annual returns. 15 Table 1 provides descriptive statistics for our sample of 128,269 observations. Initial evidence that accounting records a dollar of growth at less than a dollar is seen in the fact that the 14 See, for example, Easton et al. [2015], Gode and Ohlson [2013], Penman [2012], and Wahlen et al. [2015]. Our calculation of this variable amounts to the same calculation as seen in other texts such as Damodaran [2012] and White et al. [2003], where earnings after taxes are adjusted for accruals and for capital expenditure. To see the equivalence, note that changes in accruals are in both EPAT and change in book value of net enterprise assets (ΔNEA) and capital expenditure is part of ΔNEA; the calculation, ECF = EPAT - ΔNEA removes the accruals from EPAT and the remainder is free cash flow. In our empirical analyses, EPAT and ΔNEA are calculated following the appendix to Nissim and Penman [2001], who refer to these variables as operating income and net operating assets. We use the terms enterprise profit after tax and net enterprise assets to underscore the focus on the entity (the enterprise), which is owned by the equity and debt holders. 15 As discussed in footnote 7, in un-tabulated analyses, we examine whether our comparison with extant literature depends on the use of earnings before extraordinary items vs. comprehensive net income. Accordingly, we also truncate observations in the top and bottom 1 percent of the distribution of the variables in equation (1). 16

19 mean (median) change in enterprise value as a percentage of opening enterprise value (ΔEV) is greater (0.160 (0.051) than the mean (median) change in the book value of the enterprise as a percentage of opening enterprise value (ΔNEA) (0.064 (0.036)). The mean EPAT is less than the mean ΔNEA (0.034 compared with 0.064) and the mean (median) ECF is -029 (0.002). The mean (median) R&D plus advertising (RDADV) is (0.018) of enterprise value; percent of observations have non-zero (positive) R&D and advertising. 16 Mean (median) capital expenditures are (0.051) of enterprise value. The mean (median) ratio of the market value of net financial liabilities to opening enterprise value is (0.102) and percent of sample observations exhibit negative values of net financial liabilities, indicating that these observations have net financial assets. 17 Table 2 reports Spearman correlations among key variables. We discuss some highlights. The correlation between our dependent variable, EPAT, and the dependent variable in Basu (X/P, i.e., price scaled earnings) is and the correlation between EVg and the independent variable in Basu (RET) is Thus, it is not at all surprising that the estimates of our coefficient relating EPAT to EVg are generally very similar to the estimates of the coefficient relating earnings to equity returns. As expected the correlations between EPAT and EVg and EPAT and ECF are both positive (0.418 and 0.284) and highly significant. The correlation between EVg and ECF is significantly positive (0.157); that is, in general, higher generated value change is associated with more cash outflow. We will refer back to this table when correlations among other variables become pertinent to subsequent analyses and discussions. 16 This result is not tabulated. 17 As discussed with respect to MVD above, this is equivalent to the equity holders owning the enterprise and a pile of cash (as in Microsoft, for example). It follows that calculating EV as MVE + NFL (where negative NFL implies net financial assets) remains valid. 17

20 4. Initial Empirical Analyses A basic premise of our paper is that the accounting for change in the market value of the enterprise (i.e., growth) depends upon both the sign of the growth (i.e., growth vs. contraction) and the source of the growth. To shed light on this premise we partition the sample into the six sub-samples described in section 2. Descriptive statistics and regression results for each of the six enterprise value change sub-samples are summarized in Table 3. Panel A presents descriptive statistics for the sub-samples. Panel B presents simple Spearman correlations. Panel C presents regression results from estimating equation (6). For comparison, Panel D reports the results from a regression of earnings on returns along with the number of negative return observations in each sub-sample. (i) Enterprise Value Increasing and Value Generation We begin with a comparison of the two sub-samples of observations where the enterprise value is increasing and there is value generated from the existing enterprise assets (i.e., subsamples (1) and (2)). The results for these sub-samples are summarized in the first two columns of Table 3. In sub-sample (1), with 31,349 observations, the owners have contributed to the enterprise value change via ECF inflow while in subsample (2), which has 33,060 observations, value has been distributed to the owners of the enterprise via ECF outflow. Sub-sample (1) is comprised of enterprises that are increasing in value due to both creation of value via the enterprise assets in place and cash inflow from the owners of the enterprise. Consistent with this, the median ΔEV (0.455) reported in Panel A is the highest among the six sub-samples. Sub-sample (2) enterprises have a similar median EVg (0.291 cf ), but are distributing some of the enterprise value back to owners, resulting in a lower median ΔEV of 18

21 0.226, which is still the second-highest enterprise value change among the six subsamples. Panel A also shows that, while both sub-samples with increasing enterprise value report positive median current-period enterprise profit, they are still investing in both intangibles (the median RDADV for sub-sample (1) is similar to the median for sub-sample (2), cf ) and fixed assets (the median PPE for sub-sample (1) is similar to the median for sub-sample (2), cf ). Most of the cash inflow in sub-sample (1) comes from debt holders (median ECF of and ΔNFL of 0.086) and much of the cash outflow in sub-sample (2) goes to debt holders (median cash flow of and ΔNFL of ). 18 The enterprises that are distributing cash to the owners are more than twice the size of those receiving cash from the owners (median EV of $0.207 billion cf. $0.095 billion). Turning to the simple correlations presented in Panel B, it is notable that the correlation between EPAT and EVg is not significantly different from zero (-0.004) for sub-sample (1), but positive and significant (0.256) for sub-sample (2). The correlation between EPAT and ECF for sub-sample (1) is the lowest (0.058) among the six sub-samples, while the correlation between EPAT and ECF for sub-sample (2) is the highest among the six sub-samples (0.315). A summary of the results from the estimation of regression (6) is presented in Panel C. We continue the comparison of the two sub-samples of observations where enterprise value is increasing and there is value generated from the existing enterprise assets. The coefficient on EVg for sub-sample (1) is the smallest (i.e., least positive) of the six sub-samples consistent with the notion that, in this sample where growth is most evident, accounting captures net expenses (the estimate of the coefficient relating EPAT to EVg is significantly negative, with a t- 18 This underscores our reason for studying flows from/to all owners of the enterprise rather than just flows from equity holders. 19

22 statistic of -4.03), which are associated with the generation of profits in future periods rather than in the current period. Further, note that, while the estimate of the coefficient on EVg in significantly negative for sub-sample (1) where there is cash inflow, it is significantly positive for sub-sample (2) (0.016) when there is cash outflow. The estimate of the coefficient on ECF (i.e., 0.123) in sub-sample (1) implies that accounting records, in EPAT, per dollar of ECF inflow; in other words, the effect of ECF inflow on EPAT is overstated by 12.3 cents per dollar of ECF. The estimate of the coefficient on ECF in sub-sample (2), where ECF is positive, (i.e., there is net cash outflow) is much higher than in subsample (1) (0.318 vs ); the higher coefficient in subsample (2) shows that accounting records in EPAT more of cash outflows of enterprises that are increasing in value than of cash inflows for enterprises that are increasing in value. The interpretation of these coefficients is that the difference between accounting EPAT and economic EPAT is per dollar of ECF when there is ECF inflow and when there is ECF outflow; in other words, the over-statement of the EPAT (i.e., net expense) effect of the ECF inflow is less than the overstatement of the EPAT (i.e., net profit) effect when there is ECF outflow. The characteristics of these sub-samples (see Panel A) provide some indication of the reasons for this difference. A possible explanation is the fact that, although the enterprise value is increasing in both sub-samples, the observations in sub-sample (1) have a median increase in NEA of 15.6 percent of enterprise value while those in sub-sample (2) have virtually no change in NEA (0.015); that is, much of the cash inflow is going to build enterprise assets but, not surprisingly (in light of the fact that the enterprises in sub-sample (2) are also growing), cash 20

23 outflow is not coming from sale of enterprise assets rather it is coming from EPAT of the period. 19 For comparison, Panel D of Table 3 also reports the Basu-style estimates of earnings/return coefficients for each of the six sub-samples. The coefficient of on RET in sub-sample (1) is not significantly different from zero. Similar to the coefficient on EVg, the coefficient on RET for sub-sample (1) is the smallest earnings/return coefficient among the six sub-samples, and also the only one that is not significantly positive. On the other hand, for sub-sample (2), where there is cash outflow, the estimate of the coefficient on RET is significantly positive (0.038 with a t- statistic of 3.04). The differences across sub-samples (1) and (2), which differ according to the sign of ECF is another demonstration of the importance of considering growth due to contribution to/distributions from the owners of the firm even if the analysis is, for whatever reason, focused on the relation between earnings and equity returns. (ii) Enterprise Contraction and Value Loss Next we compare the two sub-samples (5 and 6) of observations where the enterprise value is decreasing and the existing enterprise assets are losing value. In sub-sample (5), with 24,278 observations, the owners have contributed to enterprise value change via ECF inflow; while in subsample (6), with 23,300 observations, ECF outflow has been distributed to the owners of the enterprise. The results for sub-samples (5) and (6) are summarized in the last two columns of Table This is consistent with the accumulation of unconditional accounting conservatism over time. That is, EPAT is overstated (and NEA understated) due to prior accelerated depreciation of, or disallowed capitalization of, enterprise assets. 21

24 The enterprises in sub-sample (5) are contracting due to loss in value of the enterprise assets, but they continue to receive support from the owners of the enterprise via ECF inflows. These are relatively small enterprises (median EV of $0.094 billion compared with $0.145 for subsample (6)). These enterprises are, on average, the most unprofitable enterprises across all six sub-samples (median EPAT of 0.003); they have higher intangible intensity and much lower property plant and equipment than enterprises in sub-sample (6) (median RDADV of cf and mean PPE of cf ). These are growth-oriented (vs. value-oriented) enterprises with the lowest mean BTM (0.385) of any of the six sub-samples. Consistent with the contraction experienced by these enterprises, they exhibit the highest correlation between EPAT and EVg of the six sub-samples (0.416). Moving to Panel C, the estimate of the coefficient on EVg in regression (6) of for subsample (5) indicates that there is a 17.4 cent loss in the current period per dollar of internal contraction, while the estimate of this coefficient for sub-sample (6) is smaller (0.127); in other words, more of the value lost relates to current earnings for the sub-sample of enterprises for which owners are contributing cash (presumably this cash contribution leads to expenses of the period (for example R&D, advertising and capital expenditure used to stem future losses)) relative to those where the owners are removing cash. The striking feature in the comparison across the results from sub-samples (5) and (6) is the difference in the coefficients on ECF across these two samples of contracting enterprises. Net cash inflow appears to go to support EPAT of the current period (coefficient of 0.464), whereas little of net cash outflow comes from current EPAT (insignificant coefficient on ECF). 22

25 For comparison, Panel D of Table 3 also reports the earnings/return coefficient for each subsample. The coefficients of and for sub-samples (5) and (6), respectively, are very similar to each other. Consistent with the extant literature, these coefficients are significantly positive and much higher than the coefficients for sub-samples (1) and (2). Notably, the similar earnings/returns coefficients for these two sub-samples mask the differences in ECF conservatism that become apparent when the sources of enterprise value loss are considered. (iii) Enterprise Growth, Value Loss, Net Cash Inflow The results for sub-sample (3) are summarized in column (3) of Table 3. While enterprise growth (contraction) is driven by generated value increase/decrease for most enterprises, the enterprises in sub-sample (3) are experiencing value increase due to large injections of cash by the owners of the enterprise despite experiencing value loss/contraction during the fiscal year. This is an unusual situation, evidenced by the fact that sub-sample (3) is the smallest of the six sub-samples and only contains 5.81 percent of the observations in the full sample (see Table 1, Panel B). The mean levels of EV in Panel A indicate that sub-sample (3) contains, on average, the smallest enterprises in the sample, consistent with these enterprises small market capitalization serving as a contributing factor in their ability to grow the value of the enterprise by attracting additional capital despite experiencing value loss. These enterprises are relatively unprofitable (median EPAT of 0.037), they invest little in intangibles (lowest median RDADV of 0.013), and heavily in fixed assets (median PPE of 0.366). Because of their high asset tangibility, these enterprises are able to raise a high percentage of their market value from ECF inflows (most negative median ECF across all subsamples, ), resulting in large increases in leverage (median ΔNFL of 0.191). 23

26 The results from estimation of regression (6) are summarized in Panel C. The estimate of the coefficient on EVg (0.330) indicates that there is a 33.0 cent EPAT loss in the current period per dollar of generated enterprise value loss. It is also interesting to note that the annual equity return (RET) is negative (i.e., in Basu parlance, there is bad news) for most of the observations in subsample (3) (7,043 of 7,449). Consistent with prior literature (e.g., Basu 1997), the estimates of the coefficients on EVg and the earnings/return coefficient are much higher (0.330 and 0.257, respectively, with t-statistics of 8.91and 7.44) for this sub-sample than for sub-samples (1) and (2) where EVg was positive. The estimate of the coefficient on ECF (0.062) implies that there is little conservatism in the accounting for ECF for these observations. This is consistent with the high levels of PPE in these enterprises and indicates that the majority of the financing raised by these enterprises goes towards investments that are capitalized into NEA (the implied coefficient relating ΔNEA to ECF is 0.938) rather than covering expenses of the current period. These enterprises are capitalizing ECF at a slightly higher rate than those in sub-sample (1) and again we see that the capitalization rate in this sub-sample is much higher than the liquidation rate in sub-sample (2), where the implied coefficient relating ΔNEA to ECF is 68.2 cents per dollar. (iv) Enterprise Contraction, Value Generation, Net Cash Outflow The results for the analysis of sub-sample (4) are summarized in column (4) of Table 3. Similar to sub-sample (3) this subsample is comprised of enterprises where the overall growth pattern runs counter to its internal growth pattern. In this case, the enterprise is contracting despite internal growth, a somewhat rare occurrence indicated by the fact that only 6.89 percent of our observations are in sub-sample (4). Sub-sample (4) enterprises have high beginning-of- 24