The Persistence of Cash Flow Components into Future Cash Flows

Transcription

1 The Persistence of Cash Flow Components into Future Cash Flows C. S. Agnes Cheng * Securities Exchange Commission, Washington, DC University of Houston, Houston, Texas CHENGA@SEC.GOV Dana Hollie University of Houston C.T. Bauer College of Business Department of Accountancy & Taxation 334 Melcher Hall, Suite 390-F Houston, Texas dhollie@uh.edu The Securities and Exchange Commission, as a matter of policy, disclaims responsibility for any private publication or statement by any of its employees. The views expressed herein are those of the author and do not necessarily reflect the views of the Commission or of the author's colleagues upon the staff of the Commission. *Contact Author

2 The Persistence of Cash Flow Components into Future Cash Flows Abstract We examine the persistence of cash flow components in predicting future cash flows. In this study, we evaluate six cash flow components (which parallels the direct method of the cash flow statement): cash flows related to sales, cost of goods sold, operating expenses, interest, taxes, and other. Consistent with our predictions, we find that the cash flow components from various operating activities persist differentially. We find that cash related to sales, cost of goods sold, operating expenses and interest persists a great deal into future cash flows; cash related to other has lower persistence; and cash related to taxes has no persistence. We then incorporate accrual components into our persistence regression model and find that the persistence of cash flow components are generally higher than those of accruals; however, accrual components do enhance model performance. Our findings are consistent with the AICPA s and financial analysts rationale for their recommendation that the financial effects of a company s core and non-core cash flows should be distinguished. We also corroborate prior research that both direct and indirect methods of reporting cash flows can be useful for assessing future cash flows. Our findings are relevant to financial statement users and regulators who are interested in better predicting future cash flows and researchers using cash flow prediction models to measure financial reporting quality. JEL Classification: Keywords: cash flows; accruals; cash flow prediction; direct method, indirect method 1

3 The Persistence of Cash Flow Components into Future Cash Flows I. Introduction The purpose of this paper is to document the extent of the persistence of current period cash flow components into future cash flows. It is well known that predicting cash flows is important for liquidity and solvency analysis 1 and for firm valuation especially when earnings are of a lower quality. 2 Moreover, the persistence of cash flows into the future is an essential attribute for prediction. The usefulness of cash flow information has been documented by its greater persistence over accruals; however, the differential persistence of components of aggregated cash flows into future cash flows has not been thoroughly investigated. The objective of this study is two-fold. First, we address the issue of whether cash flow components, as defined, persist differentially relating to future cash flows. Second, we examine whether the components of cash flows have the potential to improve prediction model performance. Our findings shed additional light on the importance of cash flow components in predicting future cash flows and have implications for financial statement users and accounting policy making on the reporting of cash flows. The Special Committee on Financial Reporting formed by the American Institute of Certified Public Accountants (AICPA) stated that financial statements serve users as a model of a company s business and provide considerable insight into the relations between transactions and events and the financial impact of those transactions and events on the company a key goal of financial analysis (AICPA). In general, the closer the display in financial statements maps transactions and events, the more insight it provides. Hence, the AICPA recommends that firms should distinguish between the financial effects of a company s core (major or central operations) and non-core (peripheral or 1 Krishnan and Largay III (2000) discuss the importance of using direct method in estimating future cash flows for evaluating liquidity purpose. Informal discussion with financial analysts also reveals the importance of predicting future cash flows for liquidity and solvency analysis. 2 Cheng, Liu and Schaefer (1996) show that operating cash flow provides a substitutional role when earnings are of lower permanence. 2

4 incidental activities) cash flows, thereby, presenting the best possible information in which to analyze trends in a firm without the potential distortional effects of non-core activities. Not surprisingly, financial analysts are increasingly demanding detailed information on cash flows. 3,4 The AICPA defines core activities as usual or recurring operations and recurring non-operating gains and losses. 5 Core activities should generate cash flows that persist higher into the future than cash flows from non-core activities. Krishnan and Largay (2000), using a small sample, show that cash flow components reported from the direct method and estimated direct method 6 perform better than accrual components reported from the indirect method in predicting future cash flows. Barth, Cram and Nelson (2001) (hereafter referred to as BCN) show that accrual components predict future cash flows incremental to aggregated operating cash flows. Taken together, we propose that cash flow components from various operating activities should persist differentially into future cash flows, and that the inclusion of cash flow components into the cash flow prediction model should enhance prediction model performance beyond that of accrual components. 7 It is plausible to assume different levels of persistence for cash flow components derived from the current reporting system; however, the degrees of the differences call for empirical assessment. We estimate cash flow components based 3 For example, Kyle Loughlin, an analyst at Standard & Poor and head of its chemical industry team states: I would always favor more information [over] less. Transparency and clear information about the cash flow generated from core business activities is part and parcel to good credit analysis. So, if the details are made available in a timely manner, it is an important consideration, especially in this environment. (Chang, 2002) 4 In a workshop offered by debt-rating agency, the presenters stressed the importance of disaggregating the cash flows based on their operational functions and their persistence in predicting future cash flows, the most crucial information for credit analysis. They specially mentioned the importance of taxes and the difficulty of separating cash flow paid for taxes into operating and nonoperating activities. They concur that more detailed and more transparent cash flow information will be helpful for their analyses Follow the methods used in Livnat and Zarowin (1990), and Krishnan and Largay (2000), which estimates some cash flow components based on the articulation between comparative balance sheets and other statements. 7 Krishnan and Largay III use a hold-out sample to evaluate the prediction accuracy. Barth, Cram and Nelson use explanatory power (r-squares) of the in-sample prediction model to evaluate the predication performance. We are aware that an in-sample predication performance does not necessarily lead to an out-of-sample predication accuracy (Watts and Leftwich, 1977). Due to a large number of variables and large sample observations involved and the inability of researchers to control the differences in the persistence coefficients for individual observation, we use Barth et. al in-sample predication method. We believe, if in the large sample and on average, the components of cash flows possess different persistence coefficients, there is the potential for them to improve the forecast accuracy at the individual firm level. 3

5 on the current reporting system (indirect method) for a large sample of firms. We examine whether these components persist differentially into the future, and assess whether the cash flow prediction model can be improved by considering these cash flow components. Our empirical findings enhance our understanding of the behavior of cash flow components and support the demand for detailed cash flow information in accordance with the direct method. Following the typical order of revenue and expenses reported in the multiple income statement, we disaggregate cash flows into six major components cash flows from sales, cost of goods sold, operating expenses, interest, taxes, and other. 8 We examine these components to determine whether they persist differentially into the future. We then assess whether the explanatory power of the cash flow prediction model is enhanced by this disaggregation of cash flows. If activities affecting these components do not change a great deal over time, then we should see high persistence of current cash flow to future cash flows and we should not see that decomposing cash flows into different components improves cash flow prediction. We predict that cash flow activities affecting sales, cost of goods sold and operating expenses are core operating activities and should be highly persistent. Financing activities related to interest payments involve long-term financing and should be stable across time. Tax payment activities are affected by operating, non-operating and financing transactions. They are also affected by time-dependent tax strategies. Therefore; we do not expect tax payments to be highly persistent. Activities that affect our other category include onetime operating and non-operating activities, and thus should not be highly persistent. 8 The way these revenue and expenses are reported in financial statements and summarized by Compustat (the data base we collected the data) may not be the ideal way to identify differential components. For example, taxes were not separately applied to operating and financial activities, hence, cash paid for taxes will be due to both operating and non-operating activities. Moreover, income statement reports special line items but cash flow statement does not provide corresponding cash effects, hence, our other category may include one time effect from operating or non-operating activities. Our paper does not intend to investigate what is the best way to aggregate income and cash flows to generate relevant sub-totals, we simply follow what is available from the current reporting system summarized by our data base. Our findings on no persistence of cash paid for taxes imply that the current way of reporting taxes may need improvement. 4

6 Our methodology is similar to Barth, et al. (2001); however, we focus on both the coefficients and the explanatory power in regressing next-period cash flows on this year s cash flow components. We conduct pair-wise comparisons of the coefficients on the components to assess if there exist significant differences in the persistence of the components. We also evaluate if the prediction model using cash flow components perform better than the model using only aggregate cash flows. As expected, we find that the six cash flow components persist differently into future cash flows. We find that cash flows from sales, cost of goods sold, operating expenses, and interest have similar persistence and persist more than cash flows for other expenses. We find cash related to taxes has no persistence. We also find that disaggregating cash flows into these six components significantly improves performance of the prediction model. BCN find that accrual components are important in predicting future cash flows and that they are positively associated with future cash flows. 9 The reasons that these accruals are related to future cash flows are mostly due to the realization of previously recognized accruals 10 such as the purchase of more inventories to prepare for future sales or an increase in accounts receivables to be collected in the future. The persistence of these accrual components into future cash flows depends mostly on how they revert to cash in the future. On the other hand, the persistence of cash flow components also depends on the persistence of the cash flow generating abilities of current operations. Since the persistence of accruals and cash flows is affected by different activities (or affected differently by the same activities), we expect accrual components and cash flow components complement each other in predicting future cash flows. As expected, we find that cash flow components and accrual components explain future cash flows beyond the other. Our empirical findings are consistent with the AICPA s and financial analysts rationale for their recommendation that the financial effects of a 9 In predicting next-period cash flows, they find changes in accounts receivable, changes in inventory have positive coefficients and change in accounts payable have negative coefficients. 10 Their key current operating accruals include accounts receivable, accounts payable and inventory, they are expected to be converted to cash in the future. 5

7 company s core and non-core cash flows should be distinguished in the financial statements. Furthermore, we concur with prior research that both the direct and indirect methods of reporting cash flows can be useful for assessing future cash flows (e.g., Krishnan & Largay,2000). Our findings are relevant to financial statement users who are interested in better predicting a firm s future cash flows and therefore are relevant to accounting policy makers. Moreover, our findings are specifically relevant to academic research using cash flow prediction models to measure financial reporting quality. 11 The remainder of this paper is organized as follows. Section II provides a review of the related literature. Section III describes the research design. Section IV presents the sample selection criteria and discusses our empirical findings. Section V summarizes and concludes the paper. II. Relevant Prior Literature Statement of Financial Accounting concepts (SFAC) No. 1 issued by FASB (1978) describes one of the objective of financial reporting is to provide information for users to better predict future cash flows. SFAC No. 5 further claims that accrual earnings provide a better basis for assessing firms future cash flows than past cash flow information alone (FASB, 1984, para. 24). Greenberg et al. (1986) find evidence that current earnings are a better predictor of future cash flows than are current period cash flows. In contrast, Finger (1994) find that current period cash flows have more predictive ability than current period earnings in predicting next-period cash flows. However, Lorek and Willinger (1996) show that cash flow prediction is enhanced by consideration of both earnings and accrual accounting data. Dechow, Kothari and Watts (1998) develop a model which implies earnings better predict future operating cash flows than current operating cash flows and their empirical results confirm with the model prediction. These studies combined suggest that both earnings and cash flows are important determinants in cash flow prediction. 11 For example, Cohen (2004) uses the BCN cash flow prediction model because it has the highest predictive ability compared to other models. In Cohen s study, the residual from the cash flow prediction model is used to proxy for the quality of financial reporting. Our study implies that his residual is correlated with cash flow components. 6

8 For earnings to enhance cash flow predictability beyond cash flows, behavior of accruals should provide additional insights to the enhancement in prediction. Barth, Cram and Nelson (2001) confirm that accruals improve cash flow prediction beyond aggregate cash flows. They extend the Dechow et al. (1998) model by decomposing accruals into different components and show that accrual components improve prediction of future cash flows over the aggregate accruals. Two studies that have investigated cash flow components are Krishnan and Largay (2000) and Livnat and Zarowin (1990). Using data prior to Financial Accounting Standard (FAS) No. 95, which requires firms to report cash flows from(for) various activities, Livnat and Zarowin (1990) evaluate the value relevance of cash flow components based on estimates derived from balance sheet and income statement. For the components of operating cash flows, they provide estimates including cash collected from customers, cash paid for cost of goods and operating expenses, taxes paid, interest paid and other operating cash flows. 12 They find that the value relevance of operating cash flow components are different. Krishnan and Largay (2000), using a small sample of firms that provide direct method cash flows, find that direct method cash flow information (i.e. components of operating cash flows) perform better than indirect method cash flow information (i.e. earnings plus accruals) in predicting future cash flows. They also analyze a larger sample by using estimated cash collected from customers and estimated cash paid to suppliers and employees (similar to Livnat and Zarowin) and they find cash flow components enhance cash flow prediction beyond aggregate cash flows and accrual components. To sum, Barth, et al. (2001) model the differential persistence of accrual components in predicting future cash flows and document the superior model performance when current aggregate accruals are decomposed into components in predicting future operating cash flows. In their model, they assume all cash flow components persist the same into future cash flows. Livnat and Zarowin 12 They combine cash paid for cost of goods and operating expenses. For other operating cash flows, they simply use income statement data. 7

9 document the differential value-relevance of the cash flow components. Krishnan and Largay III report cash flow components enhance cash flow prediction beyond accruals components. Taken together, their results imply cash flow components persist differently in predicting future cash flows. 13 Thus, we extend prior research by directly examining the persistence of cash flow components and their potential enhancement of cash flow prediction. III. Research Design Persistence of cash flow components depends on the persistence of the activities that generate them. If activities affecting these components do not change over time, then we should observe similar persistence across all components. Usually, firms activities are classified into three categories: operating, financing, and investing. For firms to sustain their long-term operations and growth, persistent cash flows from operations are the key attribute and many research studies focus on improving forecasts of future cash flows from operations. Future cash flows from operations can be improved by financing and investing activities. The major goal of our paper is to document the average persistence of operating cash flow components with an objective in mind as to the usefulness of detailed operating cash flows in assisting cash flow forecasts. Firms choose to invest more or finance investing more can certainly affect the persistence of current cash flows. We do not investigate the direct effects from financing and investing activities; 14 we let our results reflect their indirect effects on the persistence coefficients in our model. 13 If cash flow components improve cash flow prediction over aggregate cash flows, this will occur only if the persistence differ among the components. Lipe (1986) show the value-relevance of earnings components is affected by their persistence. This implies the findings by Livnat and Zarowin can be due to different persistence of cash flow components (if correctly priced). 14 BCN include depreciation and amortization in their cash flow prediction model, we replicate their analysis. Amount of depreciation and amortization should be highly associated with firms investment level. Like BCN, we find the depreciation expense significantly explain next period s cash flow. 8

10 To understand how the components of cash flows persist into future cash flows, we focus on comparing equations (1) and (2) as described below. 15 Equation (1) constrains the coefficients on the cash flow components so they are equal. Equation (2) relaxes the constraint and disaggregates total cash flows into six components to determine whether cash flow components reflect different information related to future cash flows. A comparison of explanatory power between equations (1) and (2) will determine whether the disaggregation of cash flows improves predictive ability relative to aggregate cash flows (e.g. Barth, Cram and Nelson). A comparison of coefficients among cash flow components in equation (2) will indicate if one component has higher impact on the dependent variable than that from other components (e.g. Livnat and Zarowin). CFO t+1 = α + βcfo t + µ t (1) CFO t+1 = α + βc_sales t + βc_cogs t + βc_oe t + βc_int t + βc_tax t + βc_other t + µ t (2) Also written as: CFO t+1 = α + βσcfo t + µ t The variables are defined as follows: 16 CFO = net cash flow from operating activities less the cash flows from extraordinary items and discontinued operations reported on the statement of cash flows (#308 - #124); C_SALES = cash flows from sales, calculated as sales minus changes in accounts receivable trade (#12- #151 or #12+#302) 17 C_COGS = cash flows from cost of goods sold, calculated as cost of goods sold (exclude depreciation) minus change in inventory plus change in accounts payable (#41- #3+ #70 or #41+#303+ #70) 18 ; C_OE = cash flow from operating and administrative expenses, calculated as operating expenses (OE=#12-#41-#13) minus change in net operating working capital (NOWC= [#4 #162 (#5 15 To keep our model expression simple, we use β indicating the coefficient and µ the error term for every variable and every model, respectively. Also, consistent with prior research, we use realized future cash flows as a proxy for future cash flows. (McNichols and Wilson, 1988; Penman and Sougiannis, 1998; Aboody et al., 1999; Barth, et al., 2001). 16 Due to data availability and companies reporting patterns (most companies report under the indirect method); the cash flow components are derived from the income statement, the comparative balance sheets and cash flow statement (except taxes and interest paid are available for some firm-year observations from the cash flow statement. When they are not available, we assign values of zero). 17 We use the balance sheet data in accounts receivable. When that is not available, we use the cash flow statement data. Both Livnat and Zarowin (1990) and arishnan and Largay III(2000) report that the differences in using cash flow statement data or balance sheet data is not high. However, Hribar and Collins(2002) report significant measurement errors between cash flow statement based and balance sheet based cash flow information. We also reverse our measurement rule, that is use cash flow statement data first and use the balance sheet data when the prior is not available. Our conclusions do not alter. 18 Compustat item 303 reports changes in inventory and item 304 reports changes in accounts payable and accrued expenses from the cash flow statement. We assume accrued expense is related to operating expense and focus on accounts payable to measure cash flows related to cost of goods sold. 9

11 #34)]) excluding changes in accounts receivable trade, inventory, accounts payable, tax payable and interest payable. 19 C_INT = cash flow related to interest payment (#315): 20 C_TAX = cash flow related to tax payment (#317): 21 C_OTHER = cash flows related to other revenue/expenses items including special and extraordinary items, calculated as cash flow from operations (#308) minus all other cash flow components (i.e., cash flows related to sales, COGS, operating expenses, interest and taxes). The six cash flow components are defined as cash flows from sales (C_SALES), cost of goods sold (C_COGS), operating and administrative expenses (C_OE), interest (C_INT), taxes (C_TAX), and other expenses (C_OTHER). All of the variables are scaled by the same scalar. We have used total number of shares, beginning total assets, average total assets as the scalar. Our results are qualitatively similar; our reported results are based on average total assets. Our estimation method builds upon Livnat and Zarowin with several differences. The major difference is that we use post-fas 95 data. SFAS No. 95 allows firms to use either direct or indirect method; however, if firms use direct method, a reconciliation between direct and indirect method is also required. Most firms choose to use the indirect method and Compustat provides data items based only on the indirect method. Therefore, we have to estimate cash flows components but we use the information provided in the cash flow statement as much as possible. 22 Livnat and Zarowin estimate cash flow collected from customers by using sales reported on the income statement and add changes in accounts receivables as reported in balance sheet. We follow their method but also use reported changes in accounts receivables from the cash flow statement as an alternative 19 Operating expenses are calculated as sales (#12) minus cost of goods sold (#41) minus operating income before depreciation (#13).The net operating working capital (NOWC), is calculated as current assets (#4) minus cash (#162) minus (current liabilities (#5) minus debt in current liabilities (#34)). Changes in accounts receivable (AR) can be derived directly from the cash flow statement (-#302) or indirectly from the comparative balance sheet ( #151). Similarly, changes in inventory (Inv) and taxes payable can be derived directly from the cash flow statement (-#303, #305 respectively) or indirectly from the balance sheet ( #3, #71 respectively). Interest payable is not provided by Compustat, we estimate it using interest expense minus interest paid reported in the cash flow statement. To sum, C_OE=OE (NOWC AR Inv + AP TP + IntP). 20 We assign zero if item #315 is missing to maintain our sample size. We also run an analysis for a restricted sample by deleting observations with missing #315. Our conclusions are not altered. 21 Compustat item #317 is tax paid reported in the cash flow statement. If it is missing, we estimate it using #16 (income taxes) adjusted for changes in income taxes payable ( #71 or #305). Our results are not altered by these two alternate measures. 22 Most firms (approximately 97% of firms according to Krishna and Largay III,2000) continue to report cash flows using the indirect method because it is allegedly more practical and cost effective to do so. Therefore, we derive direct method cash flow information mostly from indirect method cash flow information. 10

12 measurement. Livnat and Zarowin estimate cash paid to suppliers, employees, etc., which we separate into cash related to cost of goods and operating expenses. We assume accounts payable and inventory accounts are related to cost of goods sold and accrued liability and other current accounts are related to operating expenses. Our separation is not perfect and it may bias downward the persistence coefficients for these two categories. Since we are interested in contrasting cash flows related to sales to cash flows related to operating expenses, cash flows related to cost of goods sold should behave more like cash flows related to sales than cash flows related to operating expenses. If we cannot find such an result, it may be due to our assumption of these accounts, which may be a limitation of the data. However, we actually find the persistence of cash flows related to cost of goods sold is more similar to cash flows related to sales than cash flows related to operating expense. Hence, we choose to maintain the separation between cash paid for cost of goods sold and for operating expenses. Livnat and Zarowin has to estimate cash paid for interest and for taxes, we use the reported cash payment for these two purposes from the cash flow statement whenever we can. For other operating cash flows, Livnat and Zarowin use income statement data only. We use reported cash flows from operation subtracting previously estimated cash flows related to other purposes. Hribar and Collins (2002) report significant measurement error using income statement and balance sheet articulation than using the reported cash flow number. In the similar vein, we believe using the post- SFAS No. 95 data, our measures should improve over the pre-sfas No. 95 measures derived in previous studies. If we define core components as those components possessing high persistence, then some of the operating cash flow components may not represent core cash flows; alternatively, some components not from core operating activities may represent core cash flows. Still, we expect that operating activities should generate more persistent cash flows than financing activities. Sales, cost of goods sold, and operating expenses (hereafter referred to as SC&O) are generally seen as being 11

13 more related to future cash flows than are the other components and should therefore persist more than the other components. Interest should contribute less to predicting future operating cash flows since interest expense is related to financing activities rather than operating activities and financing activities may not have similar persistence to operating activities. 23 Hence, we predict that cash flows related to interest should persist less than SC&O. Taxes are related to all aspects of business including both operating and non-operating activities. However, unlike other cash flow components, which are affected by managers operating, financing, and investment activities, taxes are determined mostly by tax policies and firm tax strategies, which can be quite different from the firm s other ongoing business activities. Therefore, we predict that cash flows related to taxes should persist less than SC&O. Other expenses may consist of one-time charges like restructuring and special charges that could have differing and unpredictable effects on cash flow predictability and are therefore deemed to have low persistence. The focus of our paper is to contrast the performance of equation (1) and (2) and contrast the persistence coefficients among the cash flow components. However, we are also interested in evaluating if the components of cash flows and accruals complement each other in predicting future cash flows. We first replicate Barth et al. (2001) to ensure that our results with respect to cash flow prediction are not data or time specific. We compute cash flows and accruals consistent with BCN in order to make our comparisons. 24 Equation (3), i.e. the BCN model, is written as follows: CFO t+1 = β 0 + β 1 CFO t + β 2 AR t + β 3 AP + β 4 INV t + β 5 DEPR t + β 6 OTHER t + β 7 AMORT t + µ t (3) Also written as CFO t+1 = α + βcfo t + βσacc t + µ t The variables are defined as follows: EARN = income before extraordinary items and discontinued operation (#18); AR = change in accounts receivable per the statement of cash flows (#302); AP = change in accounts payable and accrued liabilities per the statement of cash flows (#304); INV = change in inventory per the statement of cash flows (#303); 23 US GAAP requires interest paid be included in operating cash flows while many other standards including International Accounting Standards (IAS) suggest differently. 24 Similar to BCN,we also measure accruals based on articulation between income statement and balance sheet. 12

14 DEPR = depreciation expense (#103); AMORT = amortization expense (#65); OTHER = net of all other accruals, calculated as EARN (CF + AR + INV - AP DEPR AMORT). We then extend the BCN model by decomposing the aggregate cash flows into six cash flow components. Equation (4) is written as follows: CFO t+1 = α + βσcfo t + βσacc t + µ (4) Equation (3) and (4) provide evidence as to whether cash flow components are incrementally informative beyond accrual components and aggregated cash flows alone in explaining future cash flows. IV. Sample Selection and Empirical Findings We obtain all data from the 2002 Compustat Annual Industrial, Research and Full Coverage files. For comparability purposes, we use criteria similar to BCN. Our sample excludes financial service firms (SIC codes ) because the cash flow predictability model was not developed to reflect their activities and the financial statement components and disclosure requirements differ for these industries. Our time period is from 1988 to 2002, a total of 15 years. Consistent with BCN, we exclude observations with sales less than $10 million, share price less than $1, missing Compustat data, and earnings or cash flows in the extreme upper and lower one percent of their respective distributions. The sample is constrained by data availability for calculating both accrual and cash flow components. These criteria yield a sample of 20, {Insert Table 1 about here} Table 1 reports descriptive statistics for the regression variables. We compare our descriptive statistics with those of BCN, which had only 10,164 firm-year observations in comparison to our 20, 25 We also use different sample constraints by not restricting all data available for accruals and cash flow components. Our conclusions are similar. 13

15 We find that the means of our EARN, CFO, and ACC are slightly lower than BCN s (they report 0.08 for each variable). 27 However, our standard deviations for EARN and ACC are slightly higher, while our standard deviation for CFO is similar to theirs. It should be noted that all our measures of cash flow are based on inflows (positive) and outflows (negative). Hence, we have positive means and medians for C_SALES and negative means and medians for C_COGS, C_OE, C_INT, and C_TAX, since they are all expenses. C_OTHER is positive, which suggests more other sources of revenues than expenses. {Insert Table 2 about here} Table 2 reports the Pearson and Spearman correlations for all regression variables. The correlation coefficient between C_Sales and C_COGS is particularly high ( and ). 28 The correlation coefficient between EARN and CFO is and 0.541, a correlation typical for these two variables that does not generally cause problems when they are in the same model. {Insert Table 3 about here} Regression summary statistics from equation (1) are presented in Panel A of Table 3. We use a mean analysis of regression to test the significance of the coefficients. 29 Equation (1) serves as a benchmark for assessing the relative predictive ability of aggregate cash flows to cash flow components. Consistent with prior research, we find that aggregate cash flows in equation (1) are 26 BCN examines the 1987 to 1996 time period. We delete the observations for 1987 due to its lack of previous year s data for calculations. 27 BCN report their statistics using only two digits: therefore, the difference between our numbers and theirs may be affected by rounding errors. For example, we report ACC as and they report It is likely that our number has a larger magnitude than theirs since rounding up our number will lead to When we check the sum of accruals of the means as reported (i.e. ACC = AR + INV AP - DEPR AMORT + OTHER: (-0.046)), we get (a rounding error of 0.001); however, when we check the sum of accruals of BCN, we get ACC= (-0.01) =-0.06 while they report 0.04 (a rounding error of 0.02) 28 This may cause multicollinearity problems in the regression. The best way to deal with multicollinearity is to enlarge the sample size. We have a large sample and our coefficients on C_Sales and C_COGS are all significant. Therefore, we do not feel the multicollinearity is of a great concern. However, we also conduct regression analyses by combining C_Sales and C_COGS to one cash flow component our model performance decreases a bit and the coefficient of the combined variable is very similar to those from the uncombined variables. Since the high correlation between C_Sales and C_COGS does not create problems, we keep them separate. 29 To avoid the problem of cross-sectional dependence, we examine the mean coefficients from the annual regressions using Fama-MacBeth statistics that are equal to the mean of the estimated coefficients across 15 regressions divided by the standard error of the coefficients (Fama and MacBeth, 1973). Because the Fama-Macbeth statistics are based on the coefficients from the annual regressions, they are unaffected by the potentially inflated t-statistics in the annual regressions. We apply analysis of the difference in coefficients similar to previous studies, e.g. DeFond and Hung (2003). 14

16 significantly positive in the prediction equation. CFO explains 28.69% of the variation in next-period cash flows. 30 We find that the coefficient for CFO has an average of.529 with a t-statistic of This finding suggests that more than 50% of the current year s cash flows will persist into next year s cash flows. Panel B of Table 3 presents results from estimating equation (2) in which cash flows are disaggregated. Consistent with our predictions, the six components of cash flows are significant in predicting future cash flows. The adjusted R 2 increases from 28.69% for equation (1) to 31.97% for equation (2), an 11% increase in explaining its variation. 31 We also find that all six cash flow components (C_SALES, C_COGS, C_OE, C_INT, and C_OTHER) are positive and significant, except C_TAX which is significantly negative. The coefficients for C_SALES and C_COGS are nearly.5 with a t-statistic around 26. The coefficient for C_OE has a slightly higher average (.501) but a lower t-statistic (25.06). This finding suggests that the persistence of C_OE has greater variability over time than that of C_SALES and C_COGS. This is consistent with the stickiness of operating expense suggested in Anderson, Banker and Janakiraman (2003) that many types of general administrative costs are sticky and do not vary directly with revenue. Hence, its persistence may be higher than cost of goods sold but at the same time has a higher variability. The coefficients on C_INT and C_OTHER have a value of.468 and.412, respectively; C_INT has a t-statistic of 8, while C_OTHER has a t-statistic of 17. This outcome indicates C_Other and C_INT have greater variability across years than for SC&O, especially for C_INT. This finding is consistent with C_OTHER being composed of various other expenses/revenues that vary more from year to year than the core expenses. C_INT has higher variability but interestingly, it is significant and is highly persistent in predicting next-period cash 30 BCN reports 24% for the model with CFO only. This may be due to sample differences. When we analyze our sample focusing on the observations prior to 1997, we get a similar adjusted R In contrasting model performance, we conduct analysis of yearly results on adjusted R-squares in addition to pooled analysis, we also conduct a Voung test to test the model performance. Testing results from the Voung test are similar to what we conclude based on mean analysis of the adjusted R-squared. 15

17 flows. In deciding on the reporting requirement for the statement of cash flows (Statement of Financial Accounting Standards No. 95), the FASB chose to include cash flows related to interest in the operating section, while the AICPA suggests that interest be classified as a non-operating cash flow item. Our results showing a high persistence for C_INT suggest that the categorization of cash related to interest may not be an important issue. On the other hand, the coefficient and t-statistic on TAX are and -2.99, respectively. This result implies that C_TAX does not persist into the nextperiod cash flows. Two factors may affect the persistence of cash flow from taxes. First, the persistence of taxes depends on the sources of income on which taxes are levied. Since the cash flow statement does not provide taxes paid for operating and non-operating activities separately, it is difficult to estimate, based on income statement and balance sheet data, how taxes should be distributed among these activities. Second, taxes are affected by a firm s tax strategy. Firms like to defer taxes as much as possible, and the amount of taxes a company defers depends on the timing of its real transactions. The fact that the coefficient is negative suggests that firms paying high taxes this year tend to pay lower taxes next year. To determine whether the cash flow components persist differentially statistically, we use a pair-wise test of the differences in coefficients for equation 2, which are reported in Panel C of Table 3. Each row reports the differences between the corresponding values designated in the matrix. A negative value for the mean pair-wise comparison suggests that the cash flow components designated in the column heading are less persistent than the cash flow components labeled in the stub. For example, a comparison of C_SALES and C_TAX reveals a mean difference of -.679, a t-statistic of , and a p-value of < This finding suggests that current period cash flows from taxes are less persistent than current period cash flows from sales in predicting next year s cash flows. The coefficients for the comparison of C_INT, C_TAX, and C_OTHER with C_SALES, C_COGS, and C_OE are all negative. The p-value is less than for C_TAX and C_OTHER but not significant for C_INT (p-value =.77,.73, and.69, respectively). The comparisons among 16

18 C_SALES, C_COGS, and C_OE (C_INT) reveal positive (negative) but insignificant coefficients. The comparisons among C_TAX and C_OTHER reveal significant negative values. Our findings show that C_TAX is less persistent than C_OTHER and the difference between C_INT and C_OTHER is insignificant. We find that SC&O and interest persist similarly and longer than taxes and other expenses. The persistence of interest is also similar, although slightly higher, to that of other expenses. Based on these pair-wise comparison tests, we conclude that cash flows from sales, cost of goods sold, operating expenses, and interest have high persistence while cash flows from taxes and other expenses have low persistence. If the consensus is that core cash flows should include those components that persist, then it is likely that cash flow from sales, cost of goods sold, operating expenses, and interest should be classified as core components of cash flows. Regression summary statistics for equation (3) are presented in Panel A of Table 4. Equation (3) replicates BCN and serves as a robustness check. It is also used to assess the potential predictive ability of aggregate cash flows and accrual components. Our findings reveal an adjusted R 2 of 34.27%, which is consistent with BCN (35%). We find that the coefficient for CFO has an average of.592 (0.59 in BCN) with a t-statistic of This finding suggests that almost 60% of current period cash flows will persist to next year s cash flows once the effects of the accrual components are controlled for. The coefficients of the accrual components reported in Panel A have the same signs as in BCN. The magnitudes are also similar except for INV and OTHER. We (BCN) have coefficients of (0.35) and 0.44 (0.15) for INV and OTHER, which implies a smaller inventory effect and a greater effect of OTHER in our sample. AMORT is not significant in our sample. {Insert Table 4 about here} Panel B of Table 4, presents the summary statistics for equation (4). The adjusted R 2 rises from 34.7% for equation (3) to 36.3% for equation (4), an increase of almost 5% in explaining the variation in next year s cash flows. We find that all cash flow and accrual components are significant except C_TAX and AMORT. The coefficients on cash flow components have the same signs as 17

19 those reported for equation (2) in Table 3, which does not incorporate accrual components in the model. When we include accrual components, the coefficients become larger. The Fama-Macbeth t- statistics also increases for all variables except the coefficient on C_TAX, which becomes insignificant. 32 A comparison of the coefficients for the accrual components in equation (3) and (4) reveals that the magnitudes of the coefficients of AR, INV, AP, and OTHER decrease from 0.428, 0.245, and to 0.396, 0.190, and 0.102, with OTHER having the greatest decrease, approximately 30%. The coefficients of DEPR and AMORT increase from and to and This finding implies that, in contrast to long-term accruals, the significance of short-term accruals and OTHER in equation 3 is partly due to their correlations to cash flows. Panel C of Table 4 provides a pair-wise comparison test to confirm our conclusion (from Panel C of Table 3) about the relative persistence of cash flow components once the effect of accrual components is controlled for. Our findings are similar to those in Panel C of Table 3. We also contrast the persistence coefficients between the cash flow and accrual components. The persistence coefficients of cash flow components range from to (exclude taxes paid), they are significantly higher than the persistence coefficients of the working capital accrual components (the magnitudes of the coefficients range from to 0.456). {Insert Table 5 about here} Table 5 presents a pair-wise comparison test using each model s adjusted R The adjusted R 2 increases from 28.69% for equation (1) to 31.97% for equation (2), 34.27% for equation (3), and 36.33% for equation (4). The improvement from equation to equation is around 2 3% (2.86, 2.81, 32 Adding omitted variables should improve model performance; however, the impact on the coefficients of the original variables can be either positive or negative depending on whether the significance of the original variables from the original model is due to their correlation to the added omitted variables. For example, when adding the omitted variables (the accrual components) into a model (equation 1) increases the value of the original coefficients (on the cash flow components), it means that the significance of the original variables (i.e., the cash flow components) is not due to their correlation to the omitted variables (i.e., the accrual components). 33 We also conducted Voung tests for pooled regression and the model performance is significantly different between these models. 18

20 and 2.06, respectively), with corresponding t-statistics of 6.05, 4.93, and 7.21, respectively. The corresponding improvements are significant and positive. For example, the adjusted R 2 from equation (3) (i.e. the BCN model) to equation (4) increases 2.06%, which is positive and statistically significant. This result indicates that disaggregating cash flows improves performance of the cash flow prediction model significantly. Thus, we conclude that disaggregating cash flows into cash flow components has the potential to enhance cash flow prediction whether or not the effect of accrual components is controlled for. 34 V. Summary and Conclusions Consistent with our expectations, this study provides evidence that (a) cash flow components reflect different information related to future cash flows and (b) the disaggregation of cash flow components has the potential to enhance prediction model performance. Specifically, we find that cash flows from sales, cost of goods sold, operating expenses, and interest have similar persistence in predicting future cash flows, cash paid for other expense has low persistence and cash paid for taxes has no persistence. We also find that both the cash flow components and accrual components complement each other in explaining future cash flows. Our findings will be useful from several perspectives. Financial analysts have been calling for more detailed cash flow information. Our empirical results provide a benchmark for the importance of the details of cash flows in predicting future cash flows. This findings echo the demand from the financial analysts calling for more detained information. Our findings also provide a basis for policy makers in evaluating the reporting of line items for operating cash flows. Our other category is based on estimation, it includes special items and may still include core cash flow components. We believe the persistence level will be reduced even further if we can separate more correctly the core from the non-core partitions in this category. 34 As in Barth et al. (2001), our untabulated findings reveal that our inferences throughout are unaffected by using alternative dependent variables (returns and price scaled by average total assets). For returns, the adjusted R 2 increases approximately 2% when the specifications allow the coefficients on cash flow components to differ. For price, adjusted R 2 increases more than 15%. 19

21 Keeping this in mind, the significant lower persistence of this category shall provide evidence for regulators and for managers to consider reporting more detailed one-time items in their cash flow statement. Moreover, cash paid to taxes has no persistence. In the income statement, tax expenses are required to be allocated between the line items above and below income from continuing operations. Policy makers may consider applying the same rule to cash paid for taxes. 35 Also, our finding on the importance of both the cash flow and accrual components in explaining future cash flows imply that both direct and indirect method of reporting cash flows provide useful information. Our empirical findings have implications to recent accounting research studies investigating the quality of accounting information. For example, Dechow and Dechev (2002) report the importance linking accruals and operating cash flows to identify earnings quality. In their model, future operating cash flows is an important determinant and is needed to assess earnings quality. In assessing the quality of financial reporting, Cohen (2004) evaluates the ability of accounting information in predicting future cash flows. Our results can shed additional light and provide more research avenues in this line of research. Acknowledgment The authors gratefully acknowledge the comments of Mary Geddie, K. (Shiva) Sivaramakrishnan, Scott Whisenant, and workshop participants at the University of Houston, the 2004 American Accounting Association annual meeting, Douglas Hanna, reviewer for the 2005 JAAF conference and 2005 JAAF Conference participants. 35 From an informal discussion with financial analysts, we have learned that they have to go through a lot of digging trying to separate the cash related to taxes into persistent and temporary categories. 20