«Leif Atle Beisland University of Agder Dissertation submitted to the Department of Accounting, Auditing and Law at the Norwegian School of Economics ...»
Ohlson and Penman (1992) begin by regressing stock returns on aggregate earnings. They then achieve an explanatory power, measured by mean adjusted R, of 11%. Next, they disaggregate earnings into gross margin, operating expenses, depreciation, tax expenses, other income items, extraordinary items, and total dividends declared and find that the explanatory power increases to 19%. The sign of the individual estimated regression coefficients are generally correct, but the estimated coefficients are lower for those items that are problematic from an accounting measurement perspective, such as taxes and extraordinary items. Overall, Ohlson and Penman (1992) present clear evidence that disaggregation of earnings numbers significantly improves the returns-to-earnings association. Consistent with Ohlson and Penman’s findings, Carnes (2006) finds that unexpected changes in quarterly line items (i.e., accounts receivable, inventory, current liabilities, gross margin, SGA expense, and depreciation expense) affect the value of a firm’s stock. Carnes states that “[o]ne reason line items are expected to be value-relevant is because they can provide information that is useful
been shown to be important in determining firm value.” (Carnes, 2006, p. 100). Kim et al.
(2008) also disaggregate earnings into items. They report that sales-related earnings components have a much stronger impact on stock returns than sales-unrelated earnings components. Kim et al. find that the regression coefficient of the sales margin is three times the earnings response coefficients.
The conclusions of Ohlson and Penman (1992), Carnes (2006), etc. are indirectly analysed by Barth et al. (2001). They study the effect of earnings disaggregation on cash flow predictions.
Barth et al. split aggregate earnings into cash flow and major accrual items. They claim that each accrual component reflects different information relating to future cash flows, and that aggregate earnings mask this information. Their findings reveal that disaggregating earnings into cash flow and aggregate accruals significantly increases predictive ability relative to aggregate earnings. However, disaggregating accruals into major components further significantly increases predictive ability. Barth et al. report identical conclusions when as a robustness check, they replace future cash flow with stock returns as the dependent variable of their regressions. Lev et al. (2005) also test the predictive ability of cash flow and accruals.
In their out-of-sample predictions, they actually reach opposite conclusions to Barth et al. Lev et al. state that accounting accruals do not improve the prediction of future firm performance.
This conclusion holds even if accruals are split into major items.3 The finding is attributed to the difficulties of generating reliable estimates and projections in a volatile economy, as well as the frequent misuse of such estimates by managers. Barth et al. (2005), who also perform out-of-sample predictions of equity values, report that mean squared and absolute prediction errors are smallest when disaggregating earnings into cash flow and major accrual
cash flow and total accruals. They state that if concern is with errors in the tails of distributions of the equity value prediction error, then earnings should be disaggregated into cash flow and the major accrual components; otherwise, earnings should only be disaggregated into cash flow and total accruals. Pope (2005) claims that earnings components generally do not “add up” in valuation and that accrual components are especially informative about “unusual” firms.
The effect of earnings disaggregation is investigated in several contexts. For instance, Armstrong et al. (2006) study how different earnings and balance sheet items are related to firm value in the venture-backed, private equity market. Amir and Kama (2004) analyse the influence of return on common equity components on market return. They report that net profit margin is the dominant component. Callen and Segal (2004) use a variance decomposition procedure to address the relative value relevance of news of accruals, cash flow news, and expected return news in driving firm level equity returns. After splitting net income into cash flow and accrual components, they find evidence that accrual news dominates both expected-return (discount rate) news and cash flow news in driving firm-level stock returns.
Briefly summarised, prior research suggests that accounting for the sign of aggregate earnings in value relevance regressions significantly improves the return-to-earnings association. The association also improves if earnings are disaggregated into major components. The first part Lev et al. (2005) apply the following accruals items: change in working capital items minus inventory, change in inventory, depreciation and amortization, deferred taxes, and all other operating accruals.
earnings has already been taken into account. I maintain that disaggregation is relatively more important when earnings are negative than when they are positive. Negative earnings are not related to stock returns simply because they are not expected to persist. If negative earnings were expected to persist, stock investors would liquidate the firm rather than suffer from indefinite losses.5 However, even if negative earnings are unrelated to stock returns on an aggregate level, individual earnings components may contain significant amounts of valuerelevant information (compare Pope’s (2005) assertion that earnings items do not “add up” in valuation). Individual earnings components may be persistent in cases where bottom-line earnings show little or no persistency. Such persistency could potentially be revealed if earnings are disaggregated. Note, for instance, that, while price-to-earnings ratios are frequently used for quick estimates of company value, other ratios have to be used in the case of negative earnings, such as price to cash flow or price to sales ratios (compare Kim et al., 2008). Financial statements can reveal value-relevant information even in loss cases, but one may have to dig deeper. Note that earnings disaggregation can also improve the return-toearnings association for positive earnings companies. These companies may have earnings items with different valuation impacts as well. However, positive earnings are more often persistent on an aggregate level than negative earnings. Prior research has shown that positive earnings are a significant explanatory variable for both future earnings (and cash flow) and The FASB states the following about earnings disaggregation in Objectives of Financial Reporting by Business Enterprises: “Information about enterprise earnings and its components measured by accrual accounting generally provides a better indication of enterprise performance than information about current cash receipts and payments” (FASB, 1978, paragraph 44). As far as I can see, the FASB does not propose that the information content of earnings is a function of the earnings level (for instance earnings above or below zero).
Darrough and Ye (2007) and Joos and Plesko (2005) claim that losses may be persistent if some of the costs that cause the loss in reality are investments (for instance R&D expenditure). However, even if losses are expected to prevail for some time, the investors definitely expect that they will not continue indefinitely. The investments are expected to pay off eventually and the loss will turn to a profit. Thus, even if Darrough and Ye (2007) and Joos and Plesko (2005) consider longer time horizons than Hayn (1995), the liquidation option theory proposed by Hayn is equally relevant for the cases discussed by Darrough & Ye and Joos & Plesko. Losses can under no circumstances be expected to be permanent (remember that over the lifetime of a company, summed earnings are equal to sum net cash flow), and thus they are always expected to be transitory over a time period that may be very short or quite long.
from earnings disaggregation is expected to be lower for positive than for negative earnings.
Thus, I propose the following alternative hypothesis:
Hypothesis: Disaggregation of earnings information is relatively more useful for negative than for positive earnings.
My hypothesis relates to the relative usefulness of disaggregation when the sign of earnings is taken into account. Prior research suggests that the returns-to-earnings association improves when the sign of earnings is taken into account and when earnings are disaggregated into components. In other words, the explanatory power of regression analysis is expected to increase as the sign effect and the disaggregation effect are incorporated into the regression specifications. The second part of this paper analyses which of these two effects dominate. Is the relative increase in explanatory power from the consideration of the sign of earnings more pronounced than from earnings disaggregation? Is earnings disaggregation useful when the sign of earnings is taken into account? Or is it the other way around? Is the sign of earnings really important when earnings are disaggregated? These research questions are not easily answered by theory or past empirical research. Still, the questions are highly relevant for market-based accounting research if regression models are to be correctly specified. The second part of this paper presents an explorative analysis of the relationship between the sign of earnings and the earnings disaggregation effect. As I have no expectations with respect to the relative importance of the two effects, I do not propose any hypothesis for this part of the study.
the sign of earnings and the earnings aggregation level, several studies provide evidence of the partial influence on explanatory power from accounting for one of these effects. Hayn (1995) reports an adjusted R 2 of 16.9% for positive earnings compared to 9.3% for her total sample. The explanatory power for the negative earnings sample is actually 0.0! Basu (1997) regresses abnormal returns on changes in earnings. He finds that the explanatory power increases dramatically if a dummy variable for the sign of earnings is included in his regression. The increase in adjusted R 2 varies across different assumptions concerning the length of the earnings announcement period. The lowest increase in explanatory power is 44%, while the highest increase is 429%. In their study of disaggregation of earnings items, Ohlson and Penman (1992) report an adjusted R 2 of 11% when stock returns are regressed on aggregate earnings. When earnings are split into cash flow and depreciation, the explanatory power increases to 14%. However, maximal explanatory power of 19% is reached when earnings are disaggregated into the most items possible (seven items in their study; see earlier this section). Barth et al. (2001) report an explanatory power of 10% for aggregate earnings.
When earnings are split into cash flow and aggregate accruals, the adjusted R 2 increases to 12%, and it increases further to 15% when earnings are split into cash flow and accruals items. These studies confirm that both the sign effect and the disaggregation effect are potentially important in explaining value relevance.
Note that statistical models may be seriously mis-specified if they disregard the sign effect and the different relationships that earnings items have with stock returns, if these two effects are as important as suggested by prior research. Such mis-specification depresses regression coefficients and explanatory power, and highly significant explanatory variables may appear
further light on these issues.
3 Research Design and Data Sample This section presents the research design and the data sample of the study.
3.1 Research Design and Variable Definitions Value relevance is tested using regression analysis of stock returns on accounting variables.
relevance.7 The adjusted R 2 of the regressions measure the proportion of stock returns explained by earnings variables. I employ an Easton and Harris (1991) framework and regress stock returns on earnings and their associated changes. Easton and Harris show that stock returns can be theoretically seen as a function of both earnings and change in earnings (compare also the residual income model). In their empirical study, both earnings and the first difference of earnings are generally significant explanatory variables for returns. The Easton and Harris framework is extensively applied in value relevance research (see, e.g., Brimble & Hodgson, 2007; Elgers, Porter, & Emily Xu, 2008; Francis, Schipper, & Vincent, 2003;
Adjusts for the reduced degrees of freedom as more explanatory variables are included in the regression.
Brown et al. (1999) and Gu (2007) present evidence that explanatory power may be incomparable between samples. However, I am going to compare adjusted R 2 from running different regressions on constant samples.
Brown et al. and Gu primarily discuss comparisons of explanatory power over time and across countries, but papers like Hayn (1995), where the explanatory power of a positive earnings sample is compared with the explanatory power of a negative earnings sample, are also hit by their critique.
To test the hypothesis that earnings disaggregation is relatively more important for negative than for positive earnings, I apply three levels of earnings disaggregation and run the following regressions (regarding disaggregation of the Easton & Harris specification, see Ali & Lee-Seok, 2000 ; Baruch Lev & Zarowin, 1999)10:
The threshold for categorizing earnings items as extraordinary has increased over the years in Norway.
Extraordinary items appear more frequently in the former years of my sample than in the latter. See Table 7 of the Appendix. Thus, earnings items that would have been categorized as extraordinary in the first years of the sample are more likely to be viewed as ordinary in the last years of the sample.
Change in current assets – Change in cash – Change in total current liabilities + Change in interest bearing short-term debt (inclusive of first-year instalment of long-term debt).