«Leif Atle Beisland University of Agder Dissertation submitted to the Department of Accounting, Auditing and Law at the Norwegian School of Economics ...»
(e.g., Hayn, 1995, Basu, 1997, Joos and Plesko, 2005). Several studies have also proven that disaggregated earnings, i.e., earnings split into components, are more value-relevant than aggregated earnings (e.g., Ohlson and Penman, 1992, Barth et al., 2001, Carnes, 2003, Barth et al., 2005). This study presents evidence that it is relatively more useful, using the explanatory power of regression analysis as the value relevance metric, to disaggregate earnings when bottom line earnings are negative than when they are positive. The study also investigates the relative importance of the sign effect and the aggregation effect when positive and negative earnings, respectively, are pooled into one sample. I find that it is useful to account for the sign of earnings for all earnings aggregation levels and vice versa. It is generally useful to disaggregate earnings numbers even if the sign of earnings is taken into account. The sign effect dominates the aggregation effect unless earnings are highly disaggregated. In other words, the “true” value relevance of earnings information appears to be more understated if the non-linear relationship between earnings and returns is disregarded than if the information content of all earnings items is assumed to be equal.
Ohlson and Penman (1992) show that the explanatory power of return regressions increases when earnings are disaggregated into items (after controlling for reduced degrees of freedom).
The explanatory power of their regressions is 80% higher when earnings are disaggregated into seven items than when aggregated bottom line earnings are applied. The findings are consistent with Pope’s (2003) assertion that earnings components generally do not “add up” in valuation. This study begins by analysing whether disaggregation has different consequences for a positive earnings sample versus a negative earnings sample. Hayn (1995) presents evidence that positive earnings are far more value-relevant than negative earnings. In
findings to the liquidation option held by stock investors. When companies with negative earnings exist and are not liquidated, it must be the case that investors expect that the negative earnings will not persist. Positive earnings, on the other hand, are generally much more persistent. Hayn performs her study on aggregated earnings data. I test the hypothesis that even though bottom line earnings are not expected to persist when they are negative and therefore relay little relevant information to stock investors, individual earnings items may still be highly informative. For example, compare Joos and Plesko’s (2005) assertion that investors generally do not consider losses to be homogeneous, but evaluate the causes and nature of the loss to assess its long-term implications for firm value. I hypothesize that the relative usefulness of earnings disaggregation is larger for negative earnings than for positive earnings. The explanatory power of return regressions increases from 12.96% for an aggregated earnings model to 13.62% for earnings disaggregated into cash flow and accruals items when earnings are positive. The increase in explanatory power equals 5%. However, for negative earnings, the equivalent increase is 916%, from 0.64% for the aggregated earnings model to 6.50% for the disaggregated earnings model. Overall, this finding is consistent with the hypothesis that it is relatively more useful to disaggregate earnings information when earnings are negative than when they are positive.
The first part of the study indicates that both the sign of aggregate earnings and the earnings aggregation level are important factors in value relevance research. Value relevance, measured by the explanatory power of return regressions, increases both as the sign of earnings is taken into account and as earnings are disaggregated. The second part of the study asks which of these two effects is most important. There is no simple answer to this question, as empirical analysis shows that the aggregation effect and the sign effect are both extremely
practically doubles (7.61% to 13.70%) when the sign of earnings is considered. One has to disaggregate earnings into a substantial amount of earnings items to have the same effect on explanatory power if the sign of earnings is not considered. Overall, when earnings are disaggregated and the sign of earnings effect is incorporated into the regression models, explanatory power increases by more than 150% compared to a traditional aggregate earnings specification (7.61% to 19.08%). This study instructively illustrates how the value relevance of accounting information may be seriously understated if earnings components are aggregated and the different information contents of positive and negative earnings are disregarded. This general conclusion is likely to be country-independent. The analyses are, however, performed on a Norwegian data sample.
This paper is organized as follows: Section 2 presents the theoretical background of the study and develops the hypotheses to be tested. Data and research design are described in Section 3.
Empirical findings are discussed in Section 4, and Section 5 concludes.
2 Theoretical Background and Hypothesis Development Lev (1989) assesses the usefulness of accounting earnings by evaluating a large number of studies on the relationship between stock returns and accounting earnings. He finds that most studies report a remarkably low statistical association between stock returns and current earnings. The explanatory power as measured by R from regression analyses is often below 10%, and actually approaches zero in some cases. Lev concludes that, while earnings appear to be used by investors, the extent of earnings usefulness is rather limited. He claims that low information content of reported earnings and other financial variables are important explanations for the poor returns-to-earnings association. He also states that low information
principles and, in some cases, to manipulation of reported data by managers. The seemingly low returns-to-earnings association is heavily investigated in value relevance research. Many
explanations for this phenomenon are advanced in prior research (the list is not exhaustive):
low earnings persistence (Kormendi & Lipe, 1987), lack of timeliness of earnings due to strict requirements regarding objectivity and verifiability of accounting numbers (Collins, Kothari, Shanken, & Sloan, 1994), conservative accounting (Basu, 1997; Penman & Xiao-Jun, 2002), mis-specification of statistical models (W. H. Beaver, McAnally, & Stinson, 1997; Easton & Harris, 1991; Freeman & Tse, 1992; Hayn, 1995; Liu & Thomas, 2000), overly short measurement intervals for returns and earnings (Easton, Harris, & Ohlson, 1992), aggregation of earnings items (Barth, Cram, & Nelson, 2001; Bodnar & Weintrop, 1997; Kerstein & Kim, 1995; Ohlson & Penman, 1992; Ramakrishnan & Thomas, 1998; Rayburn, 1986; Thomas, 1999), etc. This paper studies how the sign of earnings (compare Hayn, 1995) interacts with earnings disaggregation (compare Barth et al., 2001) in value relevance research.
Hayn (1995) suggests that the relationship between stock returns and accounting earnings is non-linear. Specifically, she proposes that losses are more weakly associated with stock returns than profits. She argues that losses are perceived by investors as temporary because shareholders can always liquidate the firm rather than suffer from indefinite losses. Investors hold a put option on the future cash flows of the firm that, at any time, may be exercised at a price equal to the market price of the firm’s equity. Hayn’s empirical study shows that stock price movements are much more strongly linked to current profits than to current losses.
Losses actually do not appear to be at all related with contemporaneous stock price movements. Excluding loss firms from the sample results in a near tripling in ERC (earnings response coefficient) and explanatory power. In fact, her results show that the returns-to
accounting earnings are too low to be expected to recur. According to Hayn, very low earnings would make liquidating the firm a preferred alternative to perpetuating the reported earnings level. Thus, the liquidation option theory can explain the lower explanatory power of both negative earnings and small positive earnings. Dechow and Ge (2006) claim that earnings persistence will be affected by the magnitude and sign of accruals. Specifically, they report that low accrual firms have more transitory earnings than high accrual firms. They maintain that large negative accruals originate from balance sheet adjustments relating to special items, and that these negative accruals are often indicative of firms reducing assets and downsizing. Jenkins (2003) proposes that future prospects of loss firms can be analysed through a sales-based model of future normal earnings. In this model, transitory earnings items unrelated to sales revenue are disregarded.
In contrast to Hayn’s findings, Balkrishna, et al. (2007) report that losses are relatively persistent and that the probability of loss reversal declines monotonically as the history of loss extends. According to Darrough and Ye (2007), many of the loss firms are not necessarily candidates for abandonment. Instead, they are actually likely to stay in business for many years. Many firms are able to survive and receive high market valuation because of their future prospects, in spite of current losses. Their findings are attributed to the accounting system’s inability to capture R&D and other hidden assets valued by the market. Darrough and Ye state that, as the economy shifts towards more knowledge-based industries, unrecorded intangible assets generally become more important and consequently play a larger role in valuation of firms. Joos and Plesko (2005) find that when persistent losses contain R&D, investors separately value the R&D component as an asset, but value the non-R&D component as a transitory loss. They also maintain that investors generally do not consider
term implications for firm value. Joos and Plesko conclude that investors use information beyond aggregate earnings, cash flows, and accruals to distinguish between persistent losses related to financial distress and losses related to R&D investments that have the potential to generate earnings in the future.
Expensing intangibles is one aspect of accounting conservatism. Generally, the conservatism principle1 can contribute to a non-linear returns-to-earnings relationship. Basu (1997) reverses the returns-to-earnings regression and finds that the contemporaneous sensitivity of earnings to negative returns is two to six times the sensitivity of earnings to positive returns (compare also Ball & Shivakumar, 2006). Basu attributes his findings to accounting conservatism. He interprets conservatism as resulting in the reflection of “bad news” from earnings more quickly than “good news.” Basu also reports, when running a “traditional” regression of stock returns on earnings, that negative changes in earnings have systematically lower ERCs than positive changes in earnings. He argues that positive changes in earnings are more persistent than negative changes in earnings. This asymmetric persistence is another attribute of accounting conservatism. In Australia, Balkrishna et al. (2007) find that, although conservatism is generally a pervasive aspect of accounting numbers, it is more evident among loss firms. Similarly, Klein and Marquardt (2006) report that the annual percentage of losses for U.S. firms is significantly related to accounting conservatism.2 Givoly and Hayn (2000) find that conservatism in financial reporting has increased over time.
Klein and Marquardt (2006) also find that the loss percentage is related to Compustat coverage of small firms (there is a positive association between the annual percentage of losses and the rise in the percentage of small firms appearing in the Compustat database over time), real firm performance (measured by cash flows from operations), and business cycle factors. Moreover, while conservatism is generally a significant determinant of accounting losses, its overall contribution is considerably lessened when placed alongside other independent variables.
related to stock return. Empirical models may be far better specified if one accounts for this possibly non-linear relationship. However, earnings disaggregation may also improve the return-to-earnings association. Ramakrishnan and Thomas (1998) find that the stock price-toearnings association is better described by separating components of unexpected earnings and multiplying each by a different response coefficient, rather than applying a single earnings response coefficient to aggregate unexpected earnings. Rayburn (1986) splits earnings into operating cash flows and aggregate accruals and find that both are associated with stock returns. However, when disaggregating accruals into changes in working capital, depreciation, and changes in deferred taxes, only operating cash flows and changes in working capital have significant explanatory ability.