Cash flow disaggregation and the prediction of future earnings
An earlier version of this paper was presented at a workshop at Macquarie University 2007; UTS Summer Research Symposium 2008, EAA Congress Rotterdam, 2008 and AAA Annual meeting 2008, Anaheim. We are grateful for the comments and suggestions received from workshop participants and discussants and from the referee. We also acknowledge financial support from the Discipline of Accounting and the Accounting Foundation within the University of Sydney Faculty of Economics and Business as well as Tina Huynh for excellent research assistance.
Abstract
We examine the incremental information content of the components of cash flows from operations (CFO). Specifically the research question examined in this paper is whether models incorporating components of CFO to predict future earnings provide lower prediction errors than models incorporating simply net CFO. We use Australian data in this setting as all companies were required to provide information using the direct method during the sample period. We find that the cash flow components model is superior to an aggregate cash flow model in terms of explanatory power and predictive ability for future earnings; and that disclosure of non-core (core) cash flows components is (not) useful in both respects. Our results are of relevance to investors and analysts in estimating earnings forecasts, managers of firms in regulators’ domains where choice is provided with respect to the disclosure of CFO and also to regulators’ deliberations on disclosure requirements and recommendations.
1. Introduction
Cash flow information, together with the information in accruals, can assist in the prediction of future earnings (Sloan, 1996; Pfeiffer and Elgers, 1999).1 Information in the form of aggregate cash flow from operations (CFO) includes both core and non-core components which, we argue, have differential signals with respect to future earnings. Current versions of SFAS 95, IAS 7 and AASB 107 allow managers a choice of reporting CFO using the direct method or the indirect method. The Financial Accounting Standard Board (FASB) and International Accounting Standard Board (IASB) have both proposed that the direct method should be mandatory for all companies.2 Moreover, the AASB has a preference for the direct method, expressed in previous Australian standards and during the first few years after adoption of international standards (when the AASB continued to restrict choice in the Australian versions of international standards).
We provide further evidence in relation to this debate by comparing the predictive ability of forecasting models using gross inflows and outflows of CFO (disaggregated models) with models using only net cash from operations (the aggregate models). We find that disaggregated models have higher explanatory power and lower prediction errors than the aggregate models. This evidence is of relevance to managers and regulators in making decisions and recommendations relating to cash flow information, in particular decisions regarding the level of voluntary and mandatory disclosure.
Whilst major sources and uses of cash associated with investing and financing activities must be separately disclosed on the face of the cash flow statement (CFS), regulations in some countries allow cash flows associated with operating activities to be presented using either of the direct or indirect methods. Using the direct method, the major (gross) cash inflows and outflows associated with operating activities and the sum of these items (CFO) are disclosed. By contrast, the indirect approach requires disclosure of total (aggregate) CFO and reconciliation between net income and CFO on the face of the CFS.3
Financial analysts have been persistent in advocating mandatory disclosure of direct cash flow components. Investors, credit analysts and lenders also generally support direct method disclosures on the basis of their decision-making relevance, particularly in relation to evaluations of firms’ liquidity and solvency (Smith and Freeman, 1996; Jones et al., 1998).
However, FASB and IASB standards currently permit the use of either the direct or indirect methods of reporting CFO. This choice reflects views expressed by preparers in lobbying the FASB – despite the FASB encouraging preparers to use the direct method of reporting CFO, less than 3 per cent of public firms in the USA do so (Krishnan and Largay, 2000). The frequency of use of the indirect method is similar for companies applying international accounting standards (Orpurt and Zang, 2009).
Whilst the majority of countries (including USA, Canada, EU, Switzerland) had adopted a similar position to IAS 7 and provided a choice between the direct and indirect disclosure of cash flows,4 regulators in a number of countries had reached an alternate conclusion. Australia (AASB 107), China (ASBE 31), Indonesia (PSAK 2), New Zealand (FRS-10) and South Africa (GRAP 2) require, or – until full adoption of international financial reporting standards – had required, the use of the direct method to report CFO.
Significantly, as part of the joint FASB/IASB Financial Statement Presentation project, a joint task force has revisited the issue of whether to require the use of the direct or indirect methods as part of the Financial Statement Presentation joint project.5 One issue is that the use of the indirect method may conflict with the general principle in relation to netting of cash receipts and payments. The general principle recommended by the IASB (2007) was that:
[c]ash receipts and payments should not be offset (presented net) in the statement of cash flows unless there is no incremental value in the additional information provided in a gross presentation—that is, there is no benefit in a user of the financial statements knowing the two amounts; the net amount provides all of the information that is necessary.6
The key effect of the decision to allow the use of the indirect method was that users would need to estimate the major cash flows from operating activities, and that this cost would be offset (or more than offset according to the FASB) by the cost savings to firms. Preparers argued in submissions on the FASB’s original cash flow standard that their firms’ accounting systems were not designed to collect information of the type required by the direct method and it would be ‘costly’ for companies to report information using the gross method.7 This argument would seem to be moot given advances in accounting information systems since that time.
Our paper exploits the mandated use of the direct method by Australian firms until 20078, examining the incremental information content of CFO components in predicting future earnings relative to aggregate CFO (as disclosed using the indirect method). Our evidence provides an indication of the benefits of this information to investors in forecasting future performance and to the evaluation of current regulations.
Using Australian data provides two key advantages. First, there is no self-selection bias in the sample, as all Australian firms in the sample period were required to use the direct method for presenting cash flows. Second, as detailed components are disclosed using this method, we do not introduce any estimation error by trying to estimate these components for prediction purposes. These issues are further discussed in Section 2.
This study contributes in four ways. First, we add to the existing literature on cash flow disaggregation. Whereas previous research has examined the effects on returns (e.g. Clinch et al., 2002) and subsequent cash flows (e.g. Cheng and Hollie, 2008), we complete the picture by considering the relationship between disaggregated cash flows and subsequent earnings. Second, we provide information relating to the management decision to provide aggregated versus disaggregated cash flow disclosures. Presumably management makes this decision pursuant to a cost–benefit analysis. We contribute with an analysis of the benefits to users; the extra information provided by disaggregated cash flows has the potential to reduce information asymmetry in the market.
Third, we contribute to the current debate among the IASB and FASB regarding mandated cash flow disclosures. Our evidence supports the position taken by the FASB that direct method disclosures provide information relevant to users’ decisions; specifically disclosures using the indirect method seem to be inconsistent with the general principle relating to netting off, referred to above. Our evidence also supports analysts’ contentions regarding the relevance of the direct method disclosures.
Fourth, a strength of this study is that we start with a sample not subject to self-selection bias – Australia until 2007 required full disclosure of cash from operations. This enhances both the internal and external validity of our findings.
The remainder of the paper is organized as follows. Section 2 reviews the prior literature, Section 3 develops the hypotheses, Section 4 provides details on the research design, Section 5 contains a discussion of the results, and Section 6 contains the conclusions.
2. Background
Information about components of CFO may help us understand the major sources and applications of cash from operating activities. As these components are not perfectly correlated, reporting them separately may be more useful than providing them in aggregate. Prior research indicates that a disaggregation of CFO into individual components has superior explanatory power for returns relative to models that rely only on aggregate CFO (Clinch et al., 2002). There is also evidence that CFO components (Arthur and Chuang, 2008; Krishnan and Largay, 2000) or estimates of the components of cash flow (Cheng and Hollie, 2008) have incremental information content for forecasting CFO relative to models that rely on aggregate CFO only.
Given the earnings–return relation (Ball and Brown, 1968), predicting future earnings is of importance to the market.9DeFond and Hung (2003) indicate that analysts devote much more effort to producing earnings forecasts than cash flow forecasts.10 This implies that prediction of future earnings is an important issue to analysts and investors. We are unaware of any research that specifically examines the persistence of disclosed cash flow components into future earnings.
Estimates of CFO components are in fact prone to articulation problems (Bahnson et al., 1996; Ward et al., 2005). This is contrary to the FASB (1987) assertion that direct CFO components can be readily derived using balance sheets, income statements and notes to the accounts.11 Our research is thus motivated by the mixed viewpoints between researchers, preparers and users of financial statements, and regulatory bodies, with the aim of providing further evidence relevant to this dispute.
Given that the majority of research on cash flow components has focused on their ability to predict future cash flows and to explain contemporaneous returns, we extend the existing literature (as mentioned before) by examining the potential persistence of CFO components in relation to future earnings, an area largely unexplored. Our research also contributes to the literature by utilizing a complete set of cash flow data available in Australia. This is in contrast to cash flow studies conducted in the USA that utilize a limited set of operating cash flow components data which is also subject to estimation error and self-selection bias problems, as US-based firms report CFO components voluntarily.
Several research studies have specifically explored the differential information content contained in the two alternative CFS presentation formats. In general, these studies indicate that the direct method contains more decision-relevant information for investors in predicting future firm performance and in equity valuation.
Livnat and Zarowin (1990) was the first study since the inception of SFAS No. 95 to examine the information content of individual CFO components in relation to annual security returns. They recognized that previous studies have explored the information content of cash flow only as a summary measure, constraining the components of CFO to have identical coefficients, and overlooking possible explanatory power gains through CFO disaggregation (Rayburn, 1986; Bowen et al., 1987). Using an annualized cross-sectional regression model, all operating cash flow components were found to have a strong association with security returns, except for tax payments which seem to be irrelevant to investors. Furthermore, the coefficients on components of CFO were significantly different from each other, indicating that individual components have information content beyond net CFO in explaining securities returns, and in the process impart indications that the direct method of disclosure contains more value-relevant information for investors.
Similarly, Clinch et al. (2002) investigated the extent to which components of CFO reflect information summarized in share returns. Using a data set of CFO components disclosed directly by Australian firms, analysis showed that disclosed CFO components have significantly greater explanatory power for annual returns beyond aggregate CFO, when they also have significant incremental predictive power for future operating cash flows. This is further suggestive of greater information content contained in direct method CFS disclosures relative to CFO disclosed under the indirect approach.
Other studies aimed to provide evidence on the direct versus indirect method dispute by examining whether direct method disclosures enhance cash flow prediction (Krishnan and Largay, 2000; Cheng and Hollie, 2008). In addition, these studies are capable of identifying which CFO components persist more significantly into future firm performance than others; these findings are important to investors in assessing a firm’s prospects.
Krishnan and Largay (2000) investigated whether CFO components can better predict future CFO using a sample of firms that voluntarily produce direct method CFS. Results of mean average percentage errors (MAPE) and average ranks calculated from estimated regression models revealed that the use of CFO components leads to a more accurate prediction of 1 year ahead CFO than using earnings and accruals information alone. However, the fact that US firms voluntarily report cash flow using the direct method meant only a small proportion of companies in the population were included in the sample; this is a form of self-selection bias, which may ultimately affect the generalizability of their results.
As an extension to their study, Krishnan and Largay (2000) also found articulation problems in estimates of CFO components compared with the reported counterparts, effectively casting doubts over the FASB (1987) assertion that direct method information can be accurately determined using other financial statement information.
Despite widespread evidence in the literature that suggests articulation problems in the process of estimating CFO components (Bahnson et al., 1996; Krishnan and Largay, 2000; Ward et al., 2005), many US-based studies have continued to utilize estimated CFO components to examine the usefulness of direct method disclosures, in order to address the self-selection bias problem. Cheng and Hollie (2008) is one such example. They showed that cash flows components from various operating activities persist differentially, but more specifically, CFO components related to sales, cost of goods sold, operating expenses and interest persist more strongly into future cash flows than other components. Together with accruals information, Cheng and Hollie (2008) concluded that both direct and indirect methods of reporting CFO can be useful in assessing future cash flows. This finding supports the accounting regulator’s recommendation that both methods should be used in preparing CFS.
A recent study by Arthur and Chuang (2008) addressed both the self-selection bias and estimation error problems inherent in Krishnan and Largay (2000) and Cheng and Hollie (2008), respectively, by utilizing actual CFO component data mandatorily disclosed by Australian firms. Their aim was to provide evidence on the usefulness of direct method cash flow disclosures by using a modification of the regression model presented in Cheng and Hollie (2008). Consistent with prior research, evidence indicated that CFO components perform significantly better than aggregate CFO with respect to cash flow forecasting.
The general consensus from the existing literature is that direct method disclosure significantly enhances the usefulness of cash flow information, both in cash flow prediction and in equity valuation. Our study, on the other hand, assesses how CFO disclosure choice can affect investors’ informativeness about future earnings rather than future cash flows, which would also provide evidence relevant to the direct versus indirect debate.
3. Hypotheses
The FASB contends that users of accounting information should utilize a firm’s accounting earnings as a good indication of an enterprise’s future operating cash flows.12 However, the reverse may also hold, that is, this period’s cash flows may be associated with future earnings. For example, revenues received in advance are cash inflows recognized in CFO but not included in earnings until they are ‘earned’ in subsequent periods. Similarly, prepaid expenses are a part of cash outflows but are not recognized in earnings until they are expensed in future periods. Thus, while accruals contain information about future cash flows, CFO may also embed information about future earnings performance.
In addition, Sloan (1996) proposed that accruals are less likely to persist into future earnings due to their transitory nature, whereas a higher proportion of earnings attributable to CFO signify higher ‘quality of income’, that will more likely persist into future periods. Evidence confirms that CFO is an important indicator of future earnings, in that the CFO component of earnings has a stronger relationship with future earnings (i.e. higher persistence) than the accrual component (Sloan, 1996; Barth et al., 1999; Pfeiffer and Elgers, 1999).
To date, these studies are limited to examining the usefulness of net operating cash flows in earnings forecasts. This paper extends research into the usefulness of CFO in earnings prediction by reporting evidence on the persistence of individual CFO components rather than CFO as a summary measure.
()
()On the basis of the items most commonly disclosed in indirect method reconciliations, we use the following categories of accruals:
-
ΔRECEIVABLES = increase in accounts receivable;
-
ΔPAYABLES = increase accounts payable;
-
ΔINV = increase in inventory;
-
DEPN = depreciation expense;
-
AMORT = amortization expense;
-
OTHERACC = other accruals (net).
()
Note that equation/model 4 is informationally equivalent to the disclosures available to users under the indirect method of CFO presentation. This is because, of the three components – CFO, Earnings and (the set of) accruals – knowledge of any two can yield the third through direct linear transformation.
However, model 4 may also be misspecified because there is empirical evidence (referred to in Section 2) to suggest that CFO as well as accrual components have differential persistence into future firm performance.
We disaggregate net CFO into eight components.
Cash generated from operations:
Other cash from operations:
()
()The central postulate of this paper is that directly disclosing the components of CFO will yield incremental information content relative to the disclosures provided using the indirect regime for disclosing CFO. This is essentially an examination of the incremental explanatory power of model 6 over model 4. As accrual terms are available under either disclosure regimes, the key difference between the indirect and direct methods is that the former provides net CFO only as an aggregate quantity. By implication, users are constrained to implicitly give all components of CFO an equal weighting (λ2).
By contrast, when CFO components are explicitly disclosed, users have the option of weighting any component(s) differentially, if indeed this increases explanatory power in predicting future earnings. Our first hypothesis has two forms. The first form explicitly frames a test of restrictions, measured in terms of whether explanatory power is increased due to the disaggregation:
H1a: A model which includes disaggregated CFO will have a higher explanatory power for future earnings than a model using only aggregate (net) CFO.
The second form of this hypothesis is directed at testing the same postulate slightly differently. Instead of utilizing an explicit test of restrictions on coefficients (i.e. explanatory power over the whole data set), we construct restricted (indirect) and unrestricted (direct) models over subsets of our data, and then use these models to predict 1 year ahead earnings in a hold-out sub-sample.
H1b: A model which includes disaggregated CFO will have a higher predictive ability for future earnings than a model using only aggregate (net) CFO.
As noted, the fundamental difference between these hypotheses lies in the mode of testing. H1a is tested by simply fitting a relation to the data, whereas H1b is tested by estimating models and then testing their predictions on hold-out data. Thus, H1b provides a somewhat stronger test than H1a of CFO components’ usefulness for decision making.
We classify receipts from customers and payments to suppliers and employees as the ‘core’ cash flows, as they are derived directly from a firm’s income-generating (operating) activities, and are thus generally the largest components. These components have been shown to explain or account for most of the total variations in net CFO (Clinch et al., 2002), and to be highly persistent into future operating cash flows (e.g. Arthur and Chuang, 2008). Interest related cash flows, tax payments and dividends received on the other hand can be classified as ‘non-core CFO components’, as they are more closely related to financing and investing activities rather than being cash flows derived from main operating activities. The relevance of this distinction is underscored by IAS 7 (Illustrative Examples section, p. 779), where direct method Cash flows from operation activities are partitioned into two sections: the first section includes Cash receipts from customers and Cash paid to suppliers and employees, and the overall subtotal is referred to as Cash generated from operations. The second section includes non-core components – in the IAS 7 example these are Interest paid and Income taxes paid.
Taxes paid commonly relate to the taxable income of the firm in the previous accounting period. This item thus lags, rather than leads taxable earnings. While taxable income is not the same as accounting income, we still expect a lagged relationship between earnings and taxes paid. We therefore expect that the association between taxes paid and future earnings to be less direct than the association between Cash generated from operations and future earnings.
Dividends received from associated companies are not included as part of the income of the group (IAS 128, para 11), but are instead deducted from the carrying amount of the investment. It follows that the timing of cash flow associated with the dividend will differ from the timing of the revenue recognized relating to associates. The cash flow will therefore lag rather than lead the recognition of revenue. We have no hypothesis with respect to the sign of the relationship between dividends received and future earnings.
Interest received and interest payments persist less significantly (or insignificantly) into future firm performance compared with core CFO components, as they are typically made according to a preset schedule, providing relatively little incremental information with respect to future profitability (Dechow et al., 2008; Cheng and Hollie, 2008). Furthermore, interest payments are also dependent on a company’s capital structure, an aspect that has very little direct impact on the income-generating capabilities of a firm. Accordingly, we do not have a prediction with respect to the sign of the relationship between interest cash flows and future earnings.
In summary, theory as well as evidence from prior research suggests that the components of CFO have quite differing implications for a firm’s future performance. The persistence of cash flows from core operating activities is dependent on the persistence of a firm’s cash flow generating abilities from current operations, while non-core activities such as interest and tax payments are related more to a firm’s financing and investment policies, and have actually been shown to persist less or persist insignificantly into future cash flows. Further, we expect that the persistence of CFO components into future cash flows implies a link between CFO components and future earnings, as CFO itself is a (the major) component of earnings. Our second hypothesis is thus:
H2: A model which includes CFO partitioned into core and non-core components will have a higher explanatory power for future earnings than a model using only aggregate (net) CFO.
4. Research design
4.1. Testing hypotheses 1 and 2
Hypothesis 1 is concerned with whether a disaggregation of net CFO into its direct components would improve earnings prediction power. To test this hypothesis, we compare the explanatory power of model 4 with the explanatory power of model 6.
In addition to testing the overall explanatory power of each model (H1a), we assess the two respective models’ ability to predict actual earnings (H1b); we use mean absolute percentage errors (MAPE) and average ranks, similar to measures used by McDonald (1973), Cheung et al. (1997) and more recently by Krishnan and Largay (2000) to examine the ability of CFO components to predict future cash flows.
()
()4.2. Sample data
The data used for our sample was hand collected by Aspect Huntley and provided in a customized feed. All variables retain their original sign. Importantly, all financial data on the database has been standardized; this helps mitigate the problem arising from inconsistent classification of cash flow items by companies over time and also cross-sectional differences in the classification of cash flow items. Differences between companies present more of a problem than changes in the classification of cash flows as accounting standards require that cash flows are classified in a consistent manner over time.13 All variables were scaled by total assets.14
Financial institutions are excluded from the sample due to the fact that the CFO for financial institutions does not fit into the conventional operating/investing/financing categories that are more applicable to entities in retail, wholesale or manufacturing sectors. Mining and natural resource exploration companies are also excluded because of exceptional circumstances that affect their cash flows and earnings. The cash flows of mining companies are affected by highly volatile commodity prices. The future earnings of firms engaged in resource exploration are contingent on whether discoveries have been made, and are thus mainly unrelated to current earnings.15
As CFSs were first required for Australian firms in 1992, our tests use data commencing from that year. Most Australian companies adopted international accounting standards from 2006. From this date, the measurement of both earnings and accruals changed. Therefore, the sample period spans from 1992 to 2005.16 Data required for all the independent variables are obtained from annually reported financial statements with fiscal years ending 30th June (the reporting period for most Australian firms), resulting in a total of 3672 firm-year observations for earnings prediction.
For calculation of the MAPE measure for hypothesis H1b, we estimate models using 10 years of data and then use the model to estimate the cash flow for the following year. Accordingly, we estimate 1 year ahead earnings and calculate MAPE for 2003, 2004 and 2005.
5. Data analysis
5.1. Descriptive statistics
This section presents the descriptive statistics and the correlation matrix of the variables used in testing the hypotheses. Table 1 provides descriptive statistics for all variables used in the regression models. It reveals that the CGFO components – cash received from customers (CORE_RECEIPTS) and cash paid to suppliers (CORE_PAYMENTS) – have substantially larger means (1.265 and 1.250) relative to all other cash flow components, which suggests that these two components may explain most of the total variation CFO.
| Variables | Mean | Median | SD | Skewness | Kurtosis |
|---|---|---|---|---|---|
| CORE_RECEIPTS | 1.265 | 0.931 | 1.500 | 4.401 | 35.2 |
| CORE_PAYMENTS | −1.250 | −0.909 | 1.452 | −4.777 | 38.3 |
| TAX | −0.017 | −0.002 | 0.036 | −7.603 | 163.5 |
| INTPAID | −0.017 | −0.013 | 0.024 | −8.013 | 123.8 |
| INTREC | 0.010 | 0.003 | 0.018 | 5.278 | 45.9 |
| DIV | 0.002 | 0.000 | 0.009 | 11.350 | 178.2 |
| OTHER_RECEIPTS | 0.013 | 0.000 | 0.118 | 24.777 | 769.7 |
| OTHER_PAYMENTS | −0.022 | 0.000 | 0.301 | −25.400 | 706.4 |
| CFO | −0.471 | 0.054 | 1.375 | −54.607 | 3168.1 |
| ΔRECEIVABLES | 0.062 | 0.009 | 0.716 | 27.380 | 863.3 |
| ΔPAYABLES | 0.061 | 0.009 | 0.896 | 39.664 | 1867.0 |
| ΔINV | 0.025 | 0.000 | 0.293 | 32.233 | 1210.5 |
| DEPN | −0.062 | −0.027 | 1.642 | −60.519 | 3665.7 |
| AMORT | −0.028 | −0.007 | 0.379 | −58.874 | 3532.0 |
| OTHERACC | −0.147 | −0.084 | 2.158 | −50.089 | 2801.3 |
| ACCRUALS | 0.404 | −0.041 | 1.256 | 54.557 | 3163.7 |
| EARNINGS | −0.081 | 0.037 | 1.162 | 12.955 | 875.1 |
- The sample consists of listed Australian firms from 1992 to 2005 (3672 firm-year observations). EARNINGS = after-tax operating income before extraordinary items, CFO = net cash flows from operations, ACCRUALS = EARNINGS − CFO, CORE_RECEIPTS = cash collected from customers, CORE_PAYMENTS = cash paid to suppliers, TAX = income taxes paid, INTPAID = interest paid, INTREC = interest received, DIV = dividends received, OTHER_RECEIPTS = all other disclosed cash inflow components not included in the above, OTHER_PAYMENTS = all other disclosed cash outflow components not included in the above, ΔRECEIVABLES = change in trade receivables, ΔPAYABLES = change in accounts payable and accrued liabilities, ΔINV = change in inventory, DEPN = depreciation expense, AMORT = amortization expense, OTHERACC = other accruals (net).
The relatively smaller mean of −0.471 for CFO compared with the standard deviation (1.375) shows that there is substantial variation in CFO across firms and over the years. A larger negative mean for CFO (−0.471) compared with the positive mean for ACCRUALS (0.404) partly explains the negative mean for EARNINGS (−0.081), as EARNINGS = CFO + ACCRUALS.
Table 2 presents the industry breakdown of the population of firms in comparison with the sample of firms used in the study. Note that the population proportions are calculated excluding financial institutions, mining, exploration and energy sectors, consistent with the exclusions made in the sample data. The comparison reveals that no single industry dominates the sample and all industry sectors correspond closely to the proportions in the population of listed firms in Australia. The three most represented sectors in the sample are Industrials (33.3 per cent), Consumer Discretionary (24.0 per cent) and Health Care and Biotechnology (15.0 per cent).
| Sector | Sample | Population | ||
|---|---|---|---|---|
| Number | Percentage | Number | Percentage | |
| Consumer discretionary | 881 | 24.0 | 2442 | 24.1 |
| Media | 272 | 7.4 | 608 | 6.0 |
| Health care and biotechnology | 552 | 15.0 | 1251 | 12.4 |
| Infrastructure and utilities | 101 | 2.8 | 289 | 2.8 |
| Engineering | 193 | 5.3 | 441 | 4.3 |
| Industrials | 1221 | 33.3 | 3567 | 35.1 |
| Building and construction | 278 | 7.6 | 1015 | 10.0 |
| Food and household | 174 | 4.7 | 543 | 5.3 |
| All sectors | 3672 | 100 | 10 156 | 100 |
Correlations between the cash flow components and accrual variables were also calculated. As the variables tend to exhibit relatively high kurtosis, we focus on the Spearman correlations in the discussion below.
As expected, CORE_RECEIPTS and CORE_PAYMENTS are highly correlated (−0.9), which is consistent with the economic reality that higher sales (cash receipts from customers) are associated with higher cost of goods sold or cost of providing service (cash paid to suppliers) in meeting the necessary production/service requirements. The only other Spearman correlations that are in excess of 0.5 are between DEPN and CORE_RECEIPTS, and DEPN and CORE_PAYMENTS, with coefficients of −0.556 and 0.540, respectively. All other correlations are well below ±0.8.17 We also calculate VIF for all regressions and discuss these below where relevant.
The correlation between receipts from customers and payments to suppliers and employees raises the issue of whether these variables could be combined without a loss of explanatory power. In testing hypothesis 2, we combine the components of CGFO to mitigate the problems associated with the correlation of receipts from customers and payments to suppliers and employees.
5.2. Results and discussion relating to hypothesis H1a
To test whether direct CFO components improve the prediction of future earnings beyond CFO as a summary measure, we first estimate models 4 and 6 based on a pooled sample of 3672 firm-year observations spanning 1993–2005, and then re-estimate them as annualized regressions. Model 4 uses accrual components and aggregate CFO as the independent variables, while model 6 uses accrual components and CFO components as independent variables, effectively relaxing the constraint of coefficient equivalency between the CFO components. Panels A and B of Table 3 report the results of these two pooled regressions. Both models are significant at the 1 per cent level with F-statistics of 105.556 for model 4 and 84.947 for model 6. The R2 for model 4 was 16.8 per cent, whereas R2 for model 6 reached 24.5 per cent, an increase of nearly 46 per cent over the restricted model. Using the restricted least squares (RLS) method for nested models, this overall increase in explanatory power is significant at the 1 per cent level (reported in panel C) with an F-statistic of 164.19 (p < 0.01). This provides evidence that a disaggregation of CFO into direct components contributes to a significantly higher explanatory power relative to net CFO in predicting future earnings.
| Coefficients | t-statistic | |
|---|---|---|
| Panel A: Relation of accrual components and net CFO with next period earnings (restricted model) a | ||
| Intercept | −0.057 | −3.123 |
| ΔRECEIVABLES | −0.063 | −1.034 |
| ΔPAYABLES | 0.011 | 0.268 |
| ΔINV | 0.389 | 4.916 |
| DEPN | −0.347 | −4.154 |
| AMORT | 1.056 | 3.123 |
| OTHERACC | 0.085 | 3.239 |
| CFO | 0.021 | 26.460 |
| F-statistic | 105.556 | |
| R 2 | 0.168 | |
| Adjusted R2 | 0.166 | |
| Observations | 3672 | |
| Panel B: Relation of current components of accruals and CFO with next period earnings (unrestricted model)b | ||
| Intercept | −0.089 | −3.233 |
| CORE_RECEIPTS | 0.945 | 25.828 |
| CORE_PAYMENTS | −0.965 | −26.468 |
| INTREC | −1.928 | −1.825 |
| INTPAID | 2.609 | 3.293 |
| TAX | 0.801 | 1.556 |
| DIV | 2.122 | 1.167 |
| OTHER_RECEIPTS | 0.256 | 1.713 |
| OTHER_PAYMENTS | −0.975 | −16.727 |
| ΔRECEIVABLES | −0.024 | −0.411 |
| ΔPAYABLES | −0.085 | −2.111 |
| ΔINV | 0.440 | 5.718 |
| DEPN | 0.185 | 2.318 |
| AMORT | −0.247 | −0.760 |
| OTHERACC | 0.105 | 4.131 |
| F-statistic | 84.947 | |
| R 2 | 0.245 | |
| Adjusted R2 | 0.243 | |
| Observations | 3672 | |
| Panel C: RLS test of coefficient restriction | ||
| Model 4: CFO restricted model (R2 = 0.168) | ||
| Model 6: CFO unrestricted model (R2 = 0.245) | ||
| Test of coefficient restriction | ||
| F = 164.19 | ||
| p < 0.01 | ||
- aModel 4: Earningst+1 = λ0 + λ1ΔRECEIVABLESt + λ2ΔPAYABLESt + λ3ΔINVt + λ4DEPNt + λ5AMORTt + λ6OTHERACCt + λ7CFOt + et+1. bModel 6: Earningst+1 = γ0 + γ1ΔRECEIVABLESt + γ2ΔPAYABLESt + γ3ΔINVt + γ4DEPNt + γ5AMORTt + γ6OTHERACCt + γ7CORE_RECEIPTSt + γ8CORE_PAYMENTSt + γ9TAXt + γ10INTPAIDt + γ11INTRECt + γ12DIVt + γ13OTHER_RECEIPTSt + γ14OTHER_PAYMENTSt + et+1. EARNINGS = after-tax operating income before extraordinary items, CFO = net cash flows from operations, ΔRECEIVABLES = change in trade receivables, ΔPAYABLES = change in accounts payable and accrued liabilities, ΔINV = change in inventory, DEPN = depreciation expense, AMORT = amortization expense, OTHERACC = other accruals (net). CORE_RECEIPTS = cash collected from customers, CORE_PAYMENTS = cash paid to suppliers, TAX = income taxes paid, INTPAID = interest paid, INTREC = interest received, DIV = dividends received, OTHER_RECEIPTS = all other disclosed cash outflow components not included in the above, ΔRECEIVABLES = change in trade receivables, ΔPAYABLES = change in accounts payable and accrued liabilities, ΔINV = change in inventory, DEPN = depreciation expense, AMORT = amortization expense, OTHERACC = other accruals (net).
The pooled regression used above makes an assumption that all model parameters are constant (stationary) over time. Alternatively, time-specific estimation of the models would relax this assumption. Therefore, we extend the analysis by estimating the regression models on a year-by-year basis. This would detect whether the results obtained in the pooled regressions are representative over each of the sampled years, and would also strengthen the generalizability of the models’ findings to future periods. Table 4 presents the annualized regression results for the restricted model (model 4), while Table 5 provides the results for the unrestricted model (model 6). Table 6 tabulates their respective r2 values for each sampled year, and provides the results for the test of coefficient restriction.
| 1993 | 1994 | 1995 | 1996 | 1997 | 1998 | |||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Coefficient | t-stat | Coefficient | t-stat | Coefficient | t-stat | Coefficient | t-stat | Coefficient | t-stat | Coefficient | t-stat | |
| Intercept | 0.006 | 0.196 | −0.279 | −1.431 | −0.058 | −3.536 | −0.097 | −2.143 | −0.033 | −2.024 | 0.023 | 0.863 |
| ΔRECEIVABLES | 0.700 | 5.508 | −2.966 | −4.184 | 1.146 | 10.197 | 1.640 | 6.279 | 0.227 | 2.138 | 1.223 | 16.408 |
| ΔPAYABLES | −0.314 | −6.492 | 8.046 | 15.617 | −1.393 | −9.242 | −1.522 | −5.267 | −0.221 | −2.865 | −1.052 | −7.313 |
| ΔINV | 0.616 | 2.367 | −4.631 | −2.177 | 0.878 | 5.326 | 1.951 | 6.046 | 0.112 | 0.753 | −1.209 | −7.270 |
| DEPN | 0.314 | 0.463 | 7.782 | 1.523 | 2.813 | 7.223 | 3.972 | 4.190 | 1.026 | 2.499 | 1.254 | 1.922 |
| AMORT | −0.921 | −0.990 | 0.603 | 0.176 | −0.431 | −1.229 | −4.333 | −5.222 | −0.882 | −1.968 | −0.840 | −2.155 |
| OTHERACC | 0.004 | 0.294 | −0.127 | −1.595 | 0.842 | 15.146 | 0.695 | 6.411 | 0.131 | 1.999 | 0.719 | 11.056 |
| CFO | 0.078 | 5.950 | 0.031 | 16.032 | 0.462 | 12.905 | 0.112 | 6.231 | 0.326 | 15.303 | 0.159 | 28.292 |
| R 2 | 0.788 | 0.683 | 0.592 | 0.341 | 0.593 | 0.837 | ||||||
| Adjusted R2 | 0.781 | 0.672 | 0.579 | 0.323 | 0.583 | 0.833 | ||||||
| Observations | 214 | 223 | 227 | 256 | 288 | 293 | ||||||
| 1999 | 2000 | 2001 | 2002 | 2003 | 2004 | |||||||
| Coefficient | t-stat | Coefficient | t-stat | Coefficient | t-stat | Coefficient | t-stat | Coefficient | t-stat | Coefficient | t-stat | |
| Intercept | −0.109 | −1.964 | −0.085 | −4.728 | −0.010 | −0.347 | −0.099 | −6.603 | 0.018 | 0.330 | −0.193 | −2.999 |
| ΔRECEIVABLES | 1.681 | 4.883 | 0.098 | 0.959 | −0.457 | −4.567 | 0.153 | 1.577 | 0.924 | 4.110 | −0.594 | −2.820 |
| ΔPAYABLES | −2.596 | −4.933 | 0.074 | 0.403 | 0.005 | 0.028 | −0.478 | −4.626 | −1.965 | −9.116 | −1.038 | −3.999 |
| ΔINV | 2.070 | 4.853 | 0.311 | 1.415 | 0.002 | 0.014 | 0.654 | 5.007 | 0.252 | 1.148 | 1.786 | 2.555 |
| DEPN | 2.682 | 4.739 | 0.153 | 1.477 | 0.067 | 0.105 | 1.808 | 5.677 | −2.073 | −2.178 | 4.485 | 3.938 |
| AMORT | 1.021 | 0.732 | −0.607 | −1.803 | −0.861 | −2.355 | 0.511 | 2.094 | −0.216 | −0.368 | 2.464 | 3.244 |
| OTHERACC | 1.173 | 7.913 | 0.011 | 0.186 | 0.204 | 3.487 | 0.319 | 5.055 | 0.161 | 2.444 | 1.328 | 9.252 |
| CFO | 0.314 | 7.217 | 0.571 | 9.811 | 0.244 | 7.531 | 0.478 | 19.999 | 0.127 | 9.741 | 0.455 | 17.566 |
| R 2 | 0.249 | 0.299 | 0.356 | 0.550 | 0.248 | 0.574 | ||||||
| Adjusted R2 | 0.230 | 0.281 | 0.343 | 0.542 | 0.235 | 0.567 | ||||||
| Observations | 283 | 284 | 350 | 398 | 429 | 427 | ||||||
- Model 4: Earningst+1 = λ0 + [λ1ΔRECEIVABLESt + λ2ΔPAYABLESt + λ3ΔINVt + λ4DEPNt + λ5AMORTt + λ6OTHERACCt] + λ7CFOt + et+1. EARNINGS = after-tax operating income before extraordinary items, CFO = net cash flows from operations, ΔRECEIVABLES = change in trade receivables, ΔPAYABLES = change in accounts payable and accrued liabilities, ΔINV = change in inventory, DEPN = depreciation expense, AMORT = amortization expense, OTHERACC = other accruals (net).
| 1993 | 1994 | 1995 | 1996 | 1997 | 1998 | |||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Coefficient | t-stat | Coefficient | t-stat | Coefficient | t-stat | Coefficient | t-stat | Coefficient | t-stat | Coefficient | t-stat | |
| Intercept | 0.020 | 0.570 | −0.065 | −0.278 | −0.068 | −3.288 | −0.072 | −1.427 | −0.045 | −2.201 | 0.055 | 2.276 |
| CORE_RECEIPTS | 1.208 | 12.622 | 1.682 | 2.928 | 1.013 | 13.099 | 2.294 | 15.409 | 1.112 | 16.722 | 1.198 | 33.376 |
| CORE_PAYMENTS | −1.216 | −12.167 | −1.417 | −2.403 | −1.019 | −12.867 | −2.346 | −15.453 | −1.117 | −16.406 | −1.236 | −35.812 |
| INTREC | 2.427 | 2.222 | −29.958 | −3.317 | 1.521 | 2.301 | −0.146 | −0.088 | −0.194 | −0.341 | −2.350 | −2.534 |
| INTPAID | −0.658 | −0.823 | 31.331 | 3.923 | 0.067 | 0.104 | −1.569 | −1.059 | −0.032 | −0.053 | −0.474 | −0.833 |
| TAX | −1.161 | −1.661 | 2.270 | 0.467 | −0.681 | −1.330 | −2.204 | −3.070 | −0.901 | −2.012 | −1.905 | −3.778 |
| DIV | −1.501 | −1.191 | 13.082 | 0.739 | 1.895 | 1.409 | 1.095 | 0.453 | 1.165 | 1.652 | 2.586 | 2.029 |
| OTHER_RECEIPTS | 1.082 | 2.313 | −0.076 | −0.091 | 0.881 | 5.293 | 1.503 | 6.024 | 1.008 | 5.692 | 1.379 | 4.999 |
| OTHER_PAYMENTS | −0.376 | −0.728 | −6.485 | −6.073 | −1.467 | −5.813 | −0.293 | −0.721 | −0.450 | −2.145 | −1.200 | −27.986 |
| ΔRECEIVABLES | −0.039 | −0.592 | −1.900 | −2.416 | 1.144 | 9.746 | 0.744 | 6.302 | 0.521 | 5.131 | 1.526 | 25.828 |
| ΔPAYABLES | −0.101 | −2.672 | 6.956 | 13.022 | −1.231 | −8.167 | −0.911 | −3.888 | −0.493 | −6.559 | −1.114 | −10.603 |
| ΔINV | 0.742 | 4.091 | −4.327 | −1.989 | 1.145 | 6.572 | 1.231 | 4.849 | 0.408 | 2.986 | 0.448 | 3.435 |
| DEPN | −1.353 | −2.213 | 3.291 | 0.560 | 1.394 | 3.252 | 3.308 | 4.499 | 0.630 | 1.644 | 1.367 | 2.651 |
| AMORT | −1.180 | −1.526 | 3.888 | 1.127 | −0.320 | −0.933 | −2.524 | −4.031 | −0.500 | −1.223 | −1.756 | −5.808 |
| OTHERACC | −0.018 | −1.443 | −0.162 | −2.137 | 0.729 | 14.914 | 0.883 | 8.599 | 0.363 | 5.684 | 0.968 | 18.563 |
| R 2 | 0.865 | 0.728 | 0.627 | 0.657 | 0.683 | 0.918 | ||||||
| Adjusted R2 | 0.855 | 0.710 | 0.602 | 0.637 | 0.666 | 0.914 | ||||||
| Observations | 214 | 223 | 227 | 256 | 288 | 293 | ||||||
| 1999 | 2000 | 2001 | 2002 | 2003 | 2004 | |||||||
| Coefficient | t-stat | Coefficient | t-stat | Coefficient | t-stat | Coefficient | t-stat | Coefficient | t-stat | Coefficient | t-stat | |
| Intercept | −0.279 | −3.571 | 0.433 | 3.912 | −0.010 | −0.310 | −0.053 | −2.670 | −0.046 | −0.678 | −0.265 | −2.986 |
| CORE_RECEIPTS | 1.957 | 11.968 | 1.262 | 8.262 | 0.908 | 10.024 | 0.984 | 19.257 | 1.004 | 11.433 | 1.878 | 16.080 |
| CORE_PAYMENTS | −1.916 | −11.759 | −1.014 | −7.072 | −0.916 | −10.062 | −0.985 | −18.951 | −1.016 | −11.511 | −1.865 | −15.687 |
| INTREC | 11.379 | 3.497 | 0.564 | 0.179 | 2.172 | 2.184 | −0.442 | −0.556 | −1.810 | −0.586 | 7.199 | 3.035 |
| INTPAID | −0.350 | −0.138 | −53.681 | −41.948 | −1.970 | −2.421 | −1.111 | −2.178 | −8.937 | −5.621 | 1.201 | 0.440 |
| TAX | −2.249 | −1.244 | −0.968 | −0.422 | 0.721 | 1.637 | −0.408 | −1.053 | 0.872 | 0.670 | −1.503 | −1.132 |
| DIV | 0.530 | 0.088 | 7.405 | 0.955 | 2.240 | 0.727 | 2.614 | 1.857 | 5.197 | 1.058 | −1.133 | −0.195 |
| OTHER_RECEIPTS | 2.157 | 5.575 | −0.082 | −0.048 | −0.392 | −1.068 | 0.770 | 3.293 | 0.675 | 0.657 | 1.163 | 4.645 |
| OTHER_PAYMENTS | −1.947 | −11.135 | −1.278 | −4.329 | −0.898 | −8.105 | −0.977 | −13.846 | 0.929 | 1.460 | −1.574 | −2.826 |
| ΔRECEIVABLES | 1.920 | 6.561 | 1.269 | 4.856 | −0.507 | −5.846 | 0.267 | 2.882 | 1.290 | 5.882 | 0.255 | 1.229 |
| ΔPAYABLES | −2.949 | −6.605 | −2.626 | −11.930 | 0.054 | 0.349 | −0.389 | −4.112 | −1.200 | −6.649 | −2.161 | −8.636 |
| ΔINV | 2.360 | 6.491 | 5.833 | 9.202 | 0.045 | 0.315 | 0.690 | 5.761 | −0.131 | −0.678 | 2.476 | 3.652 |
| DEPN | 3.377 | 6.672 | 0.741 | 2.403 | −0.524 | −0.973 | 0.572 | 1.996 | 2.992 | 3.369 | 5.021 | 4.421 |
| AMORT | 3.448 | 2.884 | −0.062 | −0.312 | −0.566 | −1.890 | 0.614 | 2.743 | −0.624 | −1.150 | 1.612 | 2.184 |
| OTHERACC | 1.769 | 9.341 | 0.592 | 2.496 | 0.146 | 3.064 | 0.401 | 6.898 | 0.189 | 3.070 | 1.437 | 10.245 |
| R 2 | 0.493 | 0.492 | 0.584 | 0.640 | 0.372 | 0.617 | ||||||
| Adjusted R2 | 0.467 | 0.466 | 0.566 | 0.626 | 0.351 | 0.604 | ||||||
| Observations | 283 | 284 | 350 | 398 | 429 | 427 | ||||||
- Model 6: Earningst+1 = γ0 + [γ1ΔRECEIVABLESt + γ2ΔPAYABLESt + γ3ΔINVt + γ4DEPNt + γ5AMORTt + γ6OTHERACCt] + [γ7CORE_RECEIPTSt + γ8CORE_PAYMENTSt + γ9TAXt + γ10INTPAIDt + γ11INTRECt + γ12DIVt + γ13OTHER_RECEIPTSt + γ14OTHER_PAYMENTSt] + et+1. EARNINGS = after-tax operating income before extraordinary items, CORE_RECEIPTS = cash collected from customers, CORE_PAYMENTS = cash paid to suppliers, TAX = income taxes paid, INTPAID = interest paid, INTREC = interest received, DIV = dividends received, OTHER_RECEIPTS = all other disclosed cash inflow components not included in the above, OTHER_PAYMENTS = all other disclosed cash outflow components not included in the above, ΔRECEIVABLES = change in trade receivables, ΔPAYABLES = change in accounts payable and accrued liabilities, ΔINV = change in inventory, DEPN = depreciation expense, AMORT = amortization expense, OTHERACC = other accruals (net).
| 1993 | 1994 | 1995 | 1996 | 1997 | 1998 | 1999 | 2000 | 2001 | 2002 | 2003 | 2004 | Pooled | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| CFO restricted R2 | 0.788 | 0.683 | 0.592 | 0.341 | 0.593 | 0.837 | 0.249 | 0.299 | 0.356 | 0.55 | 0.248 | 0.574 | 0.168 |
| CFO unrestricted: R2 | 0.865 | 0.728 | 0.627 | 0.657 | 0.683 | 0.918 | 0.493 | 0.492 | 0.584 | 0.640 | 0.372 | 0.617 | 0.245 |
| No. observations | 214 | 223 | 227 | 256 | 288 | 293 | 283 | 284 | 350 | 398 | 429 | 427 | 3672 |
| F-statistic | 16.215 | 4.916 | 2.842 | 31.718 | 11.073 | 39.230 | 18.425 | 14.600 | 26.229 | 13.679 | 11.678 | 6.608 | 53.281 |
| p-value | 0.000 | 0.000 | 0.007 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 |
We find that the signs of some of the independent variable coefficients in both models 4 and 6 change from year to year. Consistent with this, analysis of collinearity diagnostics shows high collinearity for several variables (including ACCREV, ΔPAYABLES and ΔINV) with other variables in the years 1996, 1998 and 2000 for model 4. Similarly, there is evidence of high collinearity for several variables (mainly CORE_RECEIPTS and CORE_PAYMENTS) in model 6 in most years. As our hypotheses relate to the explanatory power and predictive ability of the models, rather than the sign of the coefficients, the presence of collinearity is not of direct concern.18
Annual regressions (not tabulated) are estimated and we find that the unrestricted specification (model 6) is superior to the restricted specification over each of the sampled years and overall in terms of overall goodness of fit, evident by the consistently higher R2 values. To determine whether the consistent higher R2 values over the years for the unrestricted model are statistically significant, we again utilize the RLS method for nested models. The results are reported in Table 6. Resulting F-statistics reveal that model 6 has statistically high (p ≤ 0.01) explanatory power in all 12 years (1993–2004).
Inspection of the relative magnitude of the R2 values also revealed patterned variations over the sampled years. For the unrestricted models, year 1993 had an R2 value of 78.8 per cent, while 2003 had an R2 value of 0.248. However, the pattern of R2 values shows no obvious trend.
5.3.Results and discussion relating to hypothesis H1b
In addition to the analysis of the R2 values, MAPE and average ranks for years 2003–2005 are used to evaluate and compare the actual predictive performance of the two alternative models. MAPE is calculated for each model by taking the mean of the absolute difference between earnings estimated by the model and the actual realized earnings for the corresponding period. The average ranks are calculated following Cheung et al. (1997), where for each observation, a rank of one will be assigned to the model yielding the smaller absolute prediction error, and a rank of two will be given otherwise. Table 7 reports the results for MAPE and average ranks estimated for earnings years 2003–2005.
| Model | 2003 | 2004 | 2005 | |||
|---|---|---|---|---|---|---|
| MAPE | Average rank | MAPE | Average rank | MAPE | Average rank | |
| Restricted model | 0.261 | 1.693 | 0.221 | 1.664 | 0.261 | 1.693 |
| Unrestricted model | 0.195 | 1.307 | 0.176 | 1.336 | 0.195 | 1.307 |
| Friedman’s Q-statistic | 45.513 | 350.844 | 547.583 | |||
| p-value | <0.01 | <0.01 | <0.01 | |||
| Observations | 2817 | 3245 | 3672 | |||
- Lower average rank indicates higher predictive ability.
Utilizing coefficients estimated using a hold-out sample technique, MAPEs are found to be consistently lower for the unrestricted model (model 6) over the three prediction periods. This indicates that using CFO components to predict future earnings results in closer estimates than using CFO as a net measure. Friedman’s (1937)Q statistics also show that the lower average ranks for the unrestricted models are significant at the 1 per cent level over all three prediction periods, further indicating superior earnings’ predictive ability for disaggregated CFO components. These results are comparable with the ones obtained by Krishnan and Largay (2000) when they utilized the same method (i.e. MAPEs and average ranks) to evaluate the usefulness of direct versus indirect cash flow data in predicting future cash flows. Together with the results from this study, direct method CFO components therefore contain incremental information content relevant to the prediction of both forms of firm performance; operating cash flows and earnings.
Table 5 reports the t-statistics for the test of significance for the estimated coefficients based on the pooled regression analysis. The coefficients for the two CGFO components are both significant at the 5 per cent level. Non-core CFO components estimated by the pooled regression do not generally appear to exhibit statistical significance for earnings prediction. The only exception to that is the interest paid variable which is found to be positively associated with future earnings. However, similar to tests of hypothesis H1a, it is necessary to ascertain the consistency, and therefore stationarity of these pooled results by further examining the coefficients using annualized regressions. Annualized regression analysis is important for inferences to be drawn regarding the generalizability of the results for future periods. Table 5 reports the results for model 6 estimated using annual data. For all years, CGFO components are significant at the 5 per cent level with signs consistent with expectation.
5.4. Results and discussion relating to hypothesis 2
The underlying premise of our research is that finer partitioning of the information sets constituting financial statements should yield better predictive ability for future performance – in our case: earnings. Hypothesis 1 yields support for the postulate that decomposing CFO into its basic components yields an enhanced association with future earnings. Hypothesis 2 restates the premise in terms of alternative partitioning systems.
Our first alternative partitioning involves decomposing CFO into all of the previously indicated non-core components, but retaining the core components (cash receipts from customers and cash payments to suppliers and employees) as a single item (CGFO, cash generated from operations). Table 8 reports the RLS comparison of this type of partial disaggregation of CFO (model 7) against the basic model (model 4) containing only aggregate CFO.19 Surprisingly, we find that this partial disaggregation yields substantially the same explanatory power as the full disaggregation investigated in hypothesis 1.20 This suggests that disaggregating into an aggregate core component (CGFO) and a set of non-core components is sufficient to enhance explanatory power, without requiring a disaggregation of CGFO into its two components.
| 1993 | 1994 | 1995 | 1996 | 1997 | 1998 | 1999 | 2000 | 2001 | 2002 | 2003 | 2004 | Pooled | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Restricted (Model 4): R2 | 0.788 | 0.683 | 0.592 | 0.341 | 0.593 | 0.837 | 0.249 | 0.299 | 0.356 | 0.550 | 0.248 | 0.574 | 0.168 |
| Unrestricted (Model 7): R2 | 0.865 | 0.724 | 0.626 | 0.652 | 0.682 | 0.915 | 0.491 | 0.491 | 0.583 | 0.640 | 0.372 | 0.617 | 0.245 |
| No. Observations | 214 | 223 | 227 | 256 | 288 | 293 | 283 | 284 | 350 | 398 | 429 | 427 | 3672 |
| F-statistic | 113.504 | 30.899 | 19.273 | 215.376 | 76.406 | 255.106 | 127.418 | 101.470 | 182.362 | 95.750 | 81.745 | 46.256 | 372.966 |
| p-value | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 | 0.000 |
- Model 4: Earningst+1 = γ0 + [γ1ΔRECEIVABLESt + γ2ΔPAYABLESt + γ3ΔINVt + γ4DEPNt + γ5AMORTt + γ6OTHERACCt] + λ2CFOt + et+1. Model 7: Earningst+1 = γ0 + [γ1ΔRECEIVABLESt + γ2ΔPAYABLESt + γ3ΔINVt + γ4DEPNt + γ5AMORTt + γ6OTHERACCt] + [γ7,8CGFO + γ9TAXt + γ10INTPAIDt + γ11INTRECt + γ12DIVt + γ13OTHER_RECEIPTSt + γ14OTHER_PAYMENTSt] + et+1. EARNINGS = after-tax operating income before extraordinary items, CFO = net cash flows from operations, CGFO = cash collected from customers less cash paid to suppliers, TAX = income taxes paid, INTPAID = interest paid, INTREC = interest received, DIV = dividends received, OTHER_RECEIPTS = all other disclosed cash inflow components not included in the above, OTHER_PAYMENTS = all other disclosed cash outflow components not included in the above, ΔRECEIVABLES = change in trade receivables, ΔPAYABLES = change in accounts payable and accrued liabilities, ΔINV = change in inventory, DEPN = depreciation expense, AMORT = amortization expense, OTHERACC = other accruals (net).
We also consider whether there is any substantial contribution to explanatory power from the non-core components of CFO, or whether merely separating CGFO from non-core components in aggregate is sufficient to enhance explanatory power over and above reporting aggregate CFO. These results (not tabulated) indicate that aggregating non-core components into one term yields R2 lower than that of model 7,21 but substantially higher than the restricted model in most years and overall.
These results lead us to conclude that the benefits of disaggregation may be achieved without a full disaggregation of CFO into detailed components. A partial disaggregation into two components (core and non-core) is likely to be sufficient to yield a significant association with subsequent earnings. A partial disaggregation into a core component, CGFO, and the set of non-core components (INTREC, INTPAID, TAX, DIV, OTHER_RECEIPTS and OTHER_PAYMENTS) is also likely to do so.
6. Summary and conclusions
The focus of this paper is on: (i) the explanatory power and (ii) the predictive ability of the components of cash flows with respect to future earnings. We decompose earnings primarily into cash and accrual components and secondly into subcomponents. We group receipts from customers and payments together as ‘core’ operating cash flows. We believe that this is the first research which directly addresses this question using actual CFO components. Rather than attempting to estimate the components of CFO, we use a data set containing mandatory firm disclosures of CFO components. This mitigates the problem of estimation error and selection bias associated with using data from reporting environments where firms can choose whether to report the components of CFO.
We seek to determine whether such a disaggregation of aggregate CFO would enhance investors’ ability to forecast future profitability. The findings in this study are overwhelmingly in support of the above predictions, consistent with past studies that have also examined the usefulness of CFO components in predicting future firm performance (e.g. Krishnan and Largay, 2000; Arthur and Chuang, 2008; Cheng and Hollie, 2008).
We have three main findings. First, we find that the disaggregation of cash flows into the lowest level subcomponents based on reported information yields a significant increase in explanatory power over a model which just uses aggregate CFO. Second, when we test the predictive ability of the disaggregated model, we find that the prediction error is significantly lower for the disaggregated model in each of the years examined. This provides more direct evidence of the decision usefulness of disaggregated information. These findings provide strong support that the disaggregation of CFO into components is crucial to obtaining more accurate earnings projections. Third, we test whether all of the disaggregated information is useful, by separating CFO into core and non-core cash flows using the classifications suggested in IAS 7. We find that reporting core CFO as one item provides essentially the same level of explanation as does reporting it as two separate receipts and payments items. However, combining the non-core items into one element yields lower explanatory power.
These findings provide strong support that disclosure of CFO components provides information relevant to earnings prediction. Note that we control for the information in accruals. We also add to the research on accruals by continuing to find significance of various accruals for future earnings even in the presence of various disclosure regimes for cash flow.
Taken together, the results from this study have a number of direct implications for accounting regulatory bodies (policy makers), market analysts, investors and preparers of financial statements (e.g. managers). Our results support the previous Australian position on cash flow, which deviated from international accounting standards. The FASB and IASB seem to be moving towards the old Australian position (FASB, 2008). Thus, we provide substantial evidence in support of the current stance by FASB to mandate the direct method of preparing CFS. More importantly, we add to mounting evidence that CFO components convey important information to investors beyond CFO as a summary measure. The IASB should consider the mandatory disclosure of direct method CFS, which would make significant contributions to the quality of financial reporting, and enhance user appreciation of operating cash flow information. Practical application of our findings would also assist market analysts in obtaining superior earnings forecasts.
In addition, our findings will also have implications for firms in accounting jurisdictions that permit the voluntary disclosure of direct method CFS. Discretionary disclosure theory suggests that proprietary costs are an important reason why firms often withhold material information, such as cash flows information, which is generally viewed as competitively sensitive. However, evidence from this study, along with prior literature, suggests that the disclosure of CFO components is beneficial to the prediction of future firm performance. This is essentially an incentive for firms to disclose such information in order to reduce information asymmetries, pre-empt costly private information acquisition, and lower their cost of raising capital (e.g. Diamond, 1985; Diamond and Verrecchia, 1991; Verrecchia, 1983, 1990), which could potentially outweigh any proprietary costs involved.
