«Leif Atle Beisland University of Agder Dissertation submitted to the Department of Accounting, Auditing and Law at the Norwegian School of Economics ...»
All variables are scaled by market value of equity at the end of year t-111.
Explanatory power is expected to increase from regression specification (1) to regression specification (3) for both positive and negative earnings. However, the hypothesis says that the relative increase in adjusted R 2 after earnings disaggregation will be larger for the negative than for the positive earnings sample. The second part of the study is an explorative analysis of the relative importance of the earnings sign effect and the earnings disaggregation effect. The sign of earnings is taken into account by introducing a dummy for negative earnings in the regressions. For the total sample, the regressions (4) to (6) are compared with regressions (1) to (3) (a dummy variable for negative earnings in the Easton & Harris
framework is used by, for instance, Francis et al., (2003)):
I actually run regressions of even more disaggregated accounting data to test the robustness of the conclusions.
However, to keep the analysis well arranged, I will focus on the results from these three aggregation levels and only briefly comment on the results from the other specification levels.
D i, t is equal to 1 if company i’s earnings in year t are negative, and is 0 otherwise. Note that both the intercept and the slopes are allowed to be dependent on the sign of earnings.
Explanatory power is expected to increase as the disaggregation level increases, i.e., as one goes from specification (1) to (3) or (4) to (6) (Ohlson & Penman, 1992, Barth et al., 2001).
However, prior research has suggested that the adjusted R 2 effect from introducing a dummy variable for negative earnings (going from specification (1) to (4), (2) to (5) or (3) to (6)) also will be substantial (Hayn, 1995, Basu, 1997). An analysis of the explanatory power of regressions (1) to (6) is expected to provide evidence of the relative importance of the earnings sign effect and the earnings disaggregation effect. The analysis will also reveal whether positive and negative earnings have different regression coefficients on a disaggregated earnings level. The definition of separate valuation relevance applied by, for instance, Stark (1997), requires that valuation coefficients (i.e., regression coefficients) for the
3.2 Data The sample consists of firms listed on the Oslo Stock Exchange. All accounting data is obtained from the Oslo Stock Exchange’s own accounting database for quoted companies.
Stock price data is collected from the Norwegian School of Economics and Business Administration’s Stock Market Database. All stock returns are adjusted for dividends, splits, etc. Stock values and returns are measured at the 30th of December of each year.12 Observations are from 1992 to 2004. In 1992, Norwegian accounting legislation was changed to introduce deferred tax liabilities and assets (an "accounting revolution", see Hope, 1999). A major tax reform was implemented at the same time. In 2005, European law required Norwegian quoted companies to report consolidated statements in accordance with International Financial Reporting Standards (IFRS). Because the introduction of IFRS may have influenced the structural relationship between stock returns and earnings numbers, I do not include IFRS observations in this study.
Consistent with prior research, financial firms are excluded from the data sample. The original data sample consists of 1,661 observations. However, one observation is lost for each company when calculating change variables for the accounting variables. One additional observation is lost when change in accruals is calculated (due to estimation of “change in change” of working capital and deferred taxes). Observations belonging to the upper or lower percentile of RET, CF, ∆CF, ACC and ∆ACC are deleted to avoid extreme observations having unreasonably large influence on the regression results. Due to a large degree of
than the theoretical maximum of 10%. The final sample size is equal to 1,372 observations.
All accounting variables are scaled by the market value of equity at the beginning of year t. In this study, observations are aggregated cross-sectionally over time. Scaling avoids spurious correlation due to size and reduces problems with heteroskedasticity (Christie, 1987). Several scale factors could have been chosen. Most researchers deflate either by average total assets or market value of equity. I use a return specification to evaluate the value relevance of accounting earnings. In a return specification, market value of equity is the scale factor of the left-hand side of the regression. Therefore, applying market value of equity as the scale factor also for the accounting variables is consistent. Easton and Summers (2003) claim that market value of equity is the true scale factor, and, thus, the natural choice when it comes to deflating variables in capital market based accounting research.
Table 1 summarises descriptions of the variables used in this study. Panel A shows the distributional characteristics of the total sample. Mean earnings equals 1.2%13 of the starting value of equity, while the median is equal to 2.3%. Mean earnings is comprised of 12.3% cash flow and –11.1% accruals. The standard deviation for earnings is less than standard deviations for both cash flow and accruals. This indicates that accruals, to a certain extent, level out cash flow fluctuations. Depreciation is by far the most important item in accruals. The change in working capital is close to zero on average, but the dispersion is wide. Thus, the variable may In fact, prices from the last actual transactions are employed for all years. Hence, market data for the most illiquid stocks might be measured a few days prior to 30 December.
Note that a mean of 1.2% is not necessarily as low as it may seem at a first glance. Mean market deflated earnings is often not very high when long time horizons are applied. In a study of the predictive ability of accounting earnings of quoted companies in the USA, Kim and Kross (2005) report mean deflated earnings of 0.7%. Their sample is drawn from the annual Compustat industrial file for the period 1973-2000 and includes more than 100,000 observations.
Table description Table 1 shows descriptive statistics for a sample of Norwegian firms from 1992 to 2004. Panels A, B and C display the mean, first quarter, median, third quarter, standard deviation, and number of observations for the total sample, the positive earnings sample, and the negative earnings sample, respectively. Panel D lists correlation coefficients for the positive (negative) earnings sample below (above) the diagonal. Coefficients in bold denote a statistical significance at a 5% level using a two sided test.
for this sample. However, the mean of 18.8% is accompanied by a standard deviation of 75.4%. Thus, the risk is substantial. Data for sales, total assets, and market value of equity are also provided in the table. Except for the fact that market value of equity is applied to scale the accounting variables, none of these variables are actually used in the empirical study. Still, they provide some indications of the distribution of company sizes in the sample. The companies are small on average. The turnover is slightly less than 4.5 billion NOK, while total assets equal 5.7 billion NOK. However, note the substantial standard deviations for these numbers. Oslo Stock Exchange is generally comprised of small companies, but some companies are considerably larger than the average.
This study focuses on the difference between positive and negative earnings. Thus, I also report descriptive statistics for these two sub-samples. 945 observations report profits, while 427 observations report losses (i.e., a loss frequency of 31%).14 Panels B and C display the statistics for the positive and the negative earnings sample, respectively. Note that the absolute values of negative earnings are larger than the absolute values of positive earnings.
On average, negative earnings companies report relatively large deficits. This is consistent with findings of previous research (W. Beaver, McNichols, & Nelson, 2007; Burgstahler & Dichev, 1997; Hayn, 1995) that there are few companies that report earnings just below zero.15 It is also evidence that several of the negative earnings companies might have implemented a “big bath” strategy. The positive earnings companies have both larger cash Two observations have earnings equal to zero. These have been added to the positive earnings sample, but do not influence any empirical results.
This finding has often been seen as evidence that companies manage earnings to report a small profit instead of a loss (Barua, Legoria, & Moffitt, 2006; Burgstahler & Dichev, 1997; Degeorge, Patel, & Zeckhauser, 1999).
Beaver et al. (2007) show that asymmetric effects of income taxes and special items for profit and loss firms contribute to a discontinuity at zero in the distribution of earnings. Specifically, they show that effective tax rates are higher for profit firms, thereby shifting profit observations to the region just above zero. However, the magnitude and frequency of negative items are greater for loss firms, thereby shifting small loss observations away from zero.
higher for negative than for positive earnings companies. Not surprisingly, positive earnings companies have a much larger stock return than negative earnings companies. In fact, the stock return of the negative earnings sample is significantly negative on average. The three size variables reveal that negative earnings companies generally are much smaller than positive earnings companies. This result is also identical to Hayn’s (1995) findings. Nontabulated results show that the means of all the earnings items (CF, ACC, ∆WC, DEP, ∆DT) are significantly different from each other in the positive and negative earnings samples. The same holds for stock return (RET). P-values are all less than 0.1%. Note that some supplementary descriptive statistics are provided in Table 7 of the Appendix. For example, the development in the proportion of companies reporting negative earnings over time is displayed in Table 7.
Panel D of Table 1 lists the correlation coefficients between the variables applied in the empirical study. The correlations are shown for both the positive and the negative earnings samples. For the positive earnings sample, there is a significant correlation between stock returns and earnings, cash flow, and accruals. Most of the individual accruals items are also statistically related to stock returns in this bivariate analysis. However, for the negative earnings sample, there seems to be low correlations between stock returns and the accounting variables. Both total earnings and total cash flow seem to be unrelated to stock returns.
Accruals are negatively correlated with stock returns both for the positive and for the negative earnings samples. As expected, the accounting variables are highly interrelated for both samples. Nevertheless, many of the accruals items are statistically unrelated to positive earnings, but are significantly associated with negative earnings. Note that cash flow is generally correlated with all other accounting variables. Accruals and cash flows are
balance out changes in cash flow and make total earnings a more stable figure than its separate components. There are 58 significant correlation coefficients in the positive earnings sample, while the equivalent total in the negative earnings sample is 49. Thus, the accounting variables are more interrelated for positive than for negative earnings.
4 Empirical Findings Section 4.1 tests the hypothesis that it is more useful to disaggregate earnings into components when earnings are negative than when they are positive. Section 4.2 discusses the relative importance of the sign effect and the disaggregation effect in value relevance research. The rest of section 4 is devoted to testing the robustness of the empirical findings.
4.1 Disaggregation in Positive and Negative Earnings Samples My hypothesis is tested by running regressions (1) to (3) separately on the positive and the negative earnings samples. The results are presented in Table 2. Panel A shows the results from the positive earnings sample. Both earnings and the change in earnings are significant16 explanatory variables for stock returns (compare Easton and Harris, 1991). These two variables are able to explain 12.96% of the variation in returns. When earnings are split into cash flow and accruals, all explanatory variables remain significant. However, the explanatory power does not increase. The adjusted R 2 is now 12.95%, a slight decrease (due to decreased Preliminary tests show that the residuals suffer from some heteroskedasticity, but that autocorrelation does not seem to be an issue. The presented t-values are computed using White-adjusted standard errors. The White estimator for variance controls for possible heteroskedasticity in the regression analyses. Coefficients are termed “significant” if they are significant on a 5% level using two-sided tests. I have also run all regressions using Newey-West standard errors. Newey-West accounts for possible autocorrelation as well as for heteroskedasticity. The t-values of the regression coefficients are hardly affected when corrections for autocorrelation are made.
further split into their components, explanatory power increases to 13.62%. Except for the change in depreciation and “the change in change” of deferred taxes, the explanatory variables are all significant.
Table 2: Value Relevance of Positive and Negative Earnings
- 203 Table description Table 2 describes the value relevance of earnings for a sample of Norwegian firms from 1992 to 2004. It summarises the regression coefficients (Coefficient), White-adjusted t-values (t-statistic), total explanatory power (adj. R2), and number of observations (n) for the positive and the negative earnings sub-samples,
respectively. Data is analysed using the following regression specifications:
Aggregate earnings specification:
RETi, t = β0 + β1EARNi, t + β2∆EARNi, t + εi, t