«Leif Atle Beisland University of Agder Dissertation submitted to the Department of Accounting, Auditing and Law at the Norwegian School of Economics ...»
where Pit is stock price, and it is measured in the end of March in year t+1. BVSit is book value of equity per share and EPSit is earnings per share for year t.3 This model is often referred to as being based on the Ohlson  valuation framework (e.g., Collins, Maydew, and Weiss , Francis and Schipper , and Lev and Zarowin ). There are some rather well-known problems with the price model specification (e.g., Brown, Kin, and Lys , Gu ). In particular it has been shown that scale effects increase the R2, and this effect increases with the scale factor’s coefficient of variation. Comparisons between samples based on R2 may be invalid if the scale factor’s coefficient of variation differs between the samples. Several measures are taken to control for these biases, including simple remedies such as scaling by the number of outstanding shares and trimming the sample. In Prices are measured three months after the end of the fiscal year to avoid hindsight bias. We make no further adjustments of our model related to market inefficiency (c.f., Aboody, Hughes, and Liu ) but assume that the market is equally (in)efficient across our sample. Also note that our measures are per share to reduce the well-known heteroskedasticity problems in these kinds of studies (Christie ).
calculate a scale-adjusted root mean squared errors (RMSE) as recommended by Gu .
Easton and Harris  show that the changes in share price (return) is a function of earnings and the change in earnings. This is illustrated by the residual income model where the market value of equity is a function of book value and the present value of residual income. Thus, a change in the market value of equity comes from a change in book value (i.e., net earnings if we assume a clean surplus) and the change in residual earnings. When earnings proxy for residual earnings the stock return is a function of earnings and the first difference of earnings. Many empirical studies show that earnings are a significant explanatory variable for stock return. Several past studies use this model specification (e.g., Easton and Harris , Alford, Leftwich, Jones, and Zmijewski , Lev and Zarowin , Francis and
Schipper [ 1999]) and we do likewise:
where Rit, is the 12-month dividend-adjusted stock return measured from the end of March in year t to the end of March in year t+1. Earnit is the net earnings and ∆Earnit is the change in net earnings from year t-1 till year t. Also return model specifications suffer from scalerelated problems. Easton and Sommers  show that the market value of equity is the true scale factor of the firm. The starting value of equity is the obvious left-hand side deflator in return model specifications and therefore we deflate Earnit and ∆Earnit by market value of equity at the beginning of year t. As mentioned the return model is primarily used as a complement to the price model. Over the years a number of researchers have discussed the relative usefulness of the two specifications without being able to say that one outperforms
Beaver, and Landsman , Gu ). We believe that the price model is somewhat better specified when addressing our research problem. However, following the suggested precautions of e.g., Kothari and Zimmerman  both models are tested in the empirical analysis.
The association between accounting information and market value is known to be non-linear.
Both negative earnings and negative equity are unrepresentative for the future (e.g., Basu , Hayn , Ramakrishnan and Thomas ). Lev and Zarowin  argue that companies in a fast changing environment report losses more frequently. We test and control for the effect of negative earnings by introducing a dummy variable in the regression models, where D is set to 1 if EPS0, but otherwise 0 (c.f., Francis, Schipper, and Vincent ).4
We transform models (1) and (2) into:
Even though our primary metric for value relevance is the explanatory power of the regression specifications, we also analyze the value relevance of each individual explanatory variable. We apply the procedure outlined in Collins, Maydew, and Weiss  to assess the variables’ incremental value relevance. Incremental value relevance is analyzed for both the price and return model specifications. We describe the procedure for the price regression, but
Specifications with sign-dependent intercept and more interaction terms are also tested, for instance :
Pi,t = a0 + a1 BVS i,t + a 2 EPS i,t + a3 ∗ D + a 4 ∗ BVS i,t * D + a5 EPS i,t * D + ε it As such changes in the specifications do not have any material effect on the empirical results we stick to the more basic and interpretable specifications.
explanatory power from the price regression, and adjusted R 1 and adjusted R 2 the explanatory power from respectively a regression of stock price on book value per share and a regression of stock price on earnings per share. The incremental value relevance from
RBVS = RTOT − R22 REPS = RTOT − R12
The value relevance common to both explanatory variables, R COM, is computed as:
RCOM = RTOT − RBVS − REPS
3.2 Data sample The sample comprises all non-financial firms quoted at the Stockholm Stock Exchange between 1979 and 2004. We end the analysis in 2004 to avoid concerns regarding the effect that a switch to IFRS has on value relevance. The data are obtained through the Trust database provided by Six Estimates. Our initial sample contains 6006 firm-year observations. We exclude firms using other local GAAPs (in total 8 firms), but retain those that apply international accounting standards.5 Since 1998 new international accounting standards have been more or less precisely translated into Swedish and adopted by the Swedish Financial Accounting Standards Council (Redovisningsrådet). When it comes to intangible assets Sweden has no tradition of capitalizing internally generated intangible assets (as it was in In 2001 this include four firms increasing to nine firms in 2004. In no single year do they constitute more than 3% of the total observations. Because the accounting framework has to be identified manually we retain them to increase replicability. However, these observations have no material effect on the results.
capitalized with a maximum economic life of 10 years but even pharmaceutical firms with large R&D investments chose to expense them straight away. Acquired goodwill could be capitalized and amortized over a maximum of 20 years, but Swedish company law suggests an economic life of no more than five years and many firms used a shorter economic life. In 2004 (the last year before the adoption of international accounting standards) the mean (median) economic life was 11 (8) years (anonymous reference).
All firms have been classified into one of twenty industries based on the nature of their operations each year. In accordance with common practice, financially oriented firms are excluded from the analysis as their accounting framework differs substantially from that of other firms. The excluded industry categories are “investment companies”, “banks and insurance companies”, “real estate” and “other financial services”. In addition, firms with odd industry classifications are disregarded (referred to as “miscellaneous” in our database). The remaining industries are split into traditional and non-traditional industries as shown in Table
1. The non-traditional industries include most firms coupled to the “new economy”, but it is not a perfect measure of it.
After trimming the highest and lowest percentile for each variable, the data set comprises 3732 observations. Overall, we have a very small number of missing observations. In total, 72.7% of the observations are firms located in traditional industries. Table 2 presents descriptive statistics for the data sample. The distributional characteristics for the total sample are found in Panel A, while Panels B and C list the distributional characteristics for the traditional industries and the non-traditional industries, respectively. Mean earnings equals 3% of market value of equity for the total sample. Median earnings are higher than the mean
Non-traditional industries 13 Industrial development 17 High-tech development 20 Services (excl consulting and IT) 21 Consulting (excl IT) 22 Information technology services 25 Pharmaceuticals and biotechnology 26 Medical technology Traditional industries 11 Industrial manufacturing 12 Consumer manufacturing 14 Raw materials and forestry 15 Trading 16 Chemicals 18 Building and construction 19 Other production 23 Transportation Excluded industries 31 Banks and Insurance 32 Other financial services 33 Real estate 34 Investment 40 Miscellaneous Table description Table 1 displays information concerning the industry categories used in the analysis of the value relevance of accounting information, based on Swedish data from the years 1979 to 2004. On the basis of the nature of the firm’s operations in the end of an accounting period it has been placed in one of twenty industry categories.
for both traditional and non-traditional firms, suggesting that the distribution is skewed to the left. It is evident that firms operating in traditional industries are more profitable than their counterparts in non-traditional industries. The mean change in earnings is positive for both sub-samples. The relative amount of shares seems to be larger in the non-traditional sector than in the traditional sector as share prices on average are higher for firms in traditional industries. The non-traditional industries report a lower book value per share and earnings per share. The dispersion is however quite large for both groups. Firms operating in traditional
Table description Panels A, B and C of Table 2 show descriptive statistics for all industries, traditional industries and non traditional industries respectively. Each panel displays the mean, first quarter, median, third quarter and standard deviations for each of the variables used in the analysis. EARN = net earnings deflated by marked value of equity at the end of t-1, ∆EARN = yearly change in net earnings, deflated by market value of equity at the end of t-1, BVS = book value of equity per share, EPS = net earnings per share, R = the dividend-adjusted stock return, P = share price, and B/M = the book value of equity divided by the market value of equity. Panel D displays the Pearson (Spearman) correlation coefficients above (below) the diagonal for all the industries. Coefficients marked in boldface denote a statistical significance at a 5% level, two-sided test.
industries had the highest stock returns in the period. Finally, we note that the mean book-tomarket ratio of firms in the non-traditional industries is substantially lower than that of firms
lower among firms in the non-traditional industries (not tabulated).
Panel D of Table 2 displays correlation matrices for the variables applied in the regression analyses. Both Pearson and Spearman coefficients are presented for the total sample. Earnings and change in earnings have a high and significant correlation coefficient. Earnings appear to be uncorrelated with returns when parametric correlation coefficients (Pearson) are employed.
However, the non-parametric correlation coefficients (Spearman) that adjust for outliers make earnings much more closely related to stock returns. As expected, share prices have a significant association with both book value per share and earnings per share.
4. Empirical Findings Table 3 contains results from tests of the first hypothesis without any consideration of nonlinearities. The table shows regression coefficients, as well as the total and incremental explanatory power from price and return regressions. We focus the analysis on the mean coefficients and explanatory power from annual regressions. However, the table also displays figures for five-year pooled regressions and a pooled regression for the complete sample period. All regressions are run for the total sample (Panel A), and the two subsamples with traditional and non-traditional industries (Panels B and C).
Panel A of Table 3 shows that the mean adjusted R2 for the price model specification is 55% for the total sample and that most coefficients are statistically significant.67 A comparison of The significance level of the mean of the regression coefficients is estimated using the Fama and MacBeth  methodology.
For price model specifications an explanatory power of 55% is low compared to findings in past U.S. studies.
Collins, Maydew, and Weiss  report an average annual explanatory power for the 1983-1993 period of