# «Leif Atle Beisland University of Agder Dissertation submitted to the Department of Accounting, Auditing and Law at the Norwegian School of Economics ...»

Further robustness checks suggest that the conclusion above may have to be moderated slightly. Again, I run several regressions of stock returns on cash flow and accruals disaggregated into more detailed components than in regressions (3) and (6). Even though some of these specifications suffer from multicollinearity, the explanatory power of the regressions generally increases as earnings are increasingly disaggregated. When using the most disaggregated model (see Table 8 in Appendix) the adjusted R 2 is equal to 15.55% when the sign of earnings is not taken into account and 19.08% when a dummy variable for negative earnings is included in the regression. The importance of the sign effect and the disaggregation effect are both substantial. The conclusion is that regardless of the aggregation level of earnings, the explanatory power of the regressions will increase if the sign of earnings is taken into account. Similarly, if the sign of earnings is taken into account, the explanatory power will generally increase if earnings are disaggregated into components. However, in order for the disaggregation effect to be as important as the sign effect, bottom line earnings must be highly disaggregated. Table 4 presents clear evidence that one has to account for the

the income statement.

4.3 Control Variables The analysis so far assumes that value relevance is a function of the sign of earnings and the earnings aggregation level. However, prior research has shown that value relevance can be a function of several firm characteristics. Company size is one example of this kind of characteristic. For example, if the positive earnings sample and the negative earnings sample differ in average company size, and one sample mainly consists of small companies while the other mainly consists of large companies, I may be measuring the effect of company size and not the earnings sign or earnings disaggregation effect. Generally speaking, if the sign or aggregation level of earnings just proxies for something else, I may be measuring the effects of these other factors. Therefore, I now add control variables that may influence value relevance to my regression analyses. Note that these results need to be interpreted carefully.

The sign or aggregation level of earnings may be proxies for one or more of the control variables. Still, the control variables may also have a direct explanatory effect on stock returns. For instance, several studies show that company size is significantly associated with stock returns (Banz, 1981; Fama & French, 1992, 1993). If the inclusion of company size in the regressions leads to a diminished difference in explanatory power, this may be due to company size acting as an explanatory variable for stock returns, and this may indicate that this effect, for some reason, is different in the two samples. In this case, value relevance may be different between the two samples even if company size seems to be balancing out the differences in explanatory power.

of factors. The factors assumed to be most important are evaluated in this section. My first control variable is company size. According to Easton and Zmijevski (1989), value relevance may be an increasing function of company size.20 SIZE is measured as the log of the market value of equity at the end of year t. The next control variable is INTANG. INTANG is a measure of the company-specific intangible asset intensity.21 It is measured as total intangible assets deflated by the total market value of equity at the beginning of year t. Several studies suggest that high intangible asset intensity reduces the value relevance of accounting numbers (Aboody & Lev, 1998; Baruch Lev & Sougiannis, 1996; Baruch Lev & Zarowin, 1999).

Specifically, these studies find that expensing intangibles renders financial reports less valuerelevant. The Norwegian GAAP allows a great deal of flexibility regarding the treatment of intangibles. Thus, the accounting based INTANG may be an imperfect measure of the “true” intangible asset intensity of companies.22 Collins and Kothari (1989) state that value relevance is a function of growth prospects. I use the book-to-market ratio (BM) as my (inverse) proxy for expected future growth.23 However, this ratio may also be considered as a control variable for accounting conservatism. Basu Collins and Kothari (1989) also find that the return-to-earnings relationship varies with firm size. They do, however, view size as a proxy for differences in information environment, for instance, risk, growth, and persistence. Higher value relevance of large firms can also be due to their smaller loss probability (Hayn, 1995).

The loss probability is obviously indirectly controlled in my empirical tests!

The intangible assets intensity is often industry dependent. Thus, INTANG may also be viewed as a control variable for industry differences in value relevance.

According to Norwegian GAAP, intangible assets are typically expensed rather than capitalized. Capitalization of intangibles occurs somewhat randomly. In principle, because INTANG is a measure of the intangibles that actually are recorded on the balance sheet, one may expect that INTANG is positively correlated with value relevance (compare Aboody & Lev, 1998; Lev & Sougiannis, 1996). However, it may also be the case that high levels of INTANG are indications that even more intangibles are expensed. In such a case, one may expect INTANG to be negatively correlated with value relevance. If annual capital expenditure related to intangible assets had been available in the data base, this would probably have been a better indicator of the intangible asset intensity.

Actually, the book-to-market ratio may be viewed as a measure of value relevance itself, i.e., the value relevance of the balance sheet. Additionally, several studies provide evidence that book-to-market ratios are significantly related to stock returns (see for instance Fama & French, 1992, 1993; Rosenberg, Reid, & Lanstein, 1985; Stattman, 1980).

numbers. Interest rate (INTEREST) and market volatility (VOL) are also applied as control variables. Collins and Kothari (1989) find a negative relationship between interest rates and value relevance, while Easton and Zmijevski (1989) propose that value relevance is negatively related to the expected rate of return (which over time is highly correlated with the level of interest rates). Dontoh et al. (2004) suggest that value relevance is an inverse function of non-information based trading activity, and I apply market volatility as a proxy for this kind of trading. The expected return on 5-year risk-free government bonds is used as my interest rate measure, while VOL is computed as the standard deviation of monthly returns on the Oslo Stock Exchange.24 My final control variable is net reported extraordinary items scaled by the market value of equity at t-1, labelled EXTRA. I have defined earnings as earnings before extraordinary items. It is still possible that extraordinary items are related to stock returns, and that the relationship is different in the positive and negative earnings samples. Descriptive statistics for all six control variables are presented in Table 10 of the Appendix.

The control variables are included in the regressions from Tables 2 and 4. However, one needs to be careful when interpreting these results. For instance, I am not focusing on how company size directly affects stock returns. I want to study the influence of company size on value relevance, i.e., how company size affects relationships between the earnings variables’ and stock returns. Disregarding the control variables’ direct effect on stock returns is important for this study. To separate the direct effect of control variables on stock returns, I apply the incremental value relevance methodology presented by Collins et al. (1997). The

**principles of incremental value relevance are as follows:**

The OSEBX index is applied. OSEBX is a value-weighted, investable index consisting of a representative selection of exchange listed companies on the Oslo Stock Exchange.

where EV = earnings var iable (EARN, CF, ACC, WC, DEP, DT − depending on aggregation level) CV = control var iable (SIZE, INTANG, BM, INTEREST, VOL, EXTRA ) The explanatory power from this regression is labelled R TOT. I then run one regression on only the earnings variables (see Table 2) and one regression on only the control variables. The explanatory power values from these regressions are labelled R 1 and R 2, respectively. The incremental explanatory power of the earnings variables and the control variables,

**respectively, can now be defined as:**

The results from the incremental explanatory power analyses are presented in Tables 5 and 6.

These tables are equivalent to Table 3 and Panel D of Table 4, respectively, with the only

explanatory power is the chosen measure of value relevance, I do not focus on individual regression coefficients in these robustness checks. It turns out that all control variables except for EXTRA generally have significant regression coefficients. Details of these regressions can be found in Tables 11 and 12 of the Appendix.

Table 5: Incremental Explanatory Power in Positive and Negative Earnings Samples Table description Table 5 lists total and incremental explanatory power (further regression details are provided in Table 11 of the Appendix) from regressions of stock returns on earnings variables and six control variables for the positive and negative earnings samples, respectively. Explanatory power is analysed for a sample of Norwegian firms from 1992 to 2004.

**R TOT is the adjusted R 2 from the following regressions:**

where RETi,t is the stock return for company i in year t, EARN is earnings before extraordinary items, CF is cash flow from operations, ACC is total accruals, WC is working capital, DEP is depreciation and impairment, and DT is deferred taxes. ∆ denotes yearly change in the variables. The accounting variables are scaled by the market value of equity on 30 December in year t-1. CVi is control variable i. The control variables are company size (log of market value of equity), intangible asset intensity (sum of intangible assets at time t divided by the market value of equity at the beginning of year t), the book-to-market ratio (book value of equity divided by market value of equity at time t), interest rate (the expected return on 5-year risk free government bonds), stock price

Define R 1 as the explanatory power from a regression that only includes the earnings variables, and R 2 as the explanatory power from a regression that only includes the control variables. The incremental value relevance of

**earnings, R 2, and the control variables, R CON, can then be defined as:**

EARN

Table description Table 6 lists total and incremental explanatory power (further regression details are provided in Table 12 of the Appendix) from regressions of stock returns on earnings variables and six control variables for a sample of

**Norwegian firms from 1992 to 2004. R TOT is the adjusted R 2 from the following regressions:**

where RETi,t is the stock return for company i in year t, EARN is earnings before extraordinary items, CF is cash flow from operations, ACC is total accruals, WC is working capital, DEP is depreciation and impairment, and DT is deferred taxes. D is a dummy variable equal to 1 when earnings are negative, 0 otherwise. ∆ denotes yearly change in the variables. The accounting variables are scaled by the market value of equity on 30 December in year t-1. CVi is control variable i. The control variables are company size (log of market value of equity), intangible asset intensity (sum of intangible assets at time t divided by the market value of equity at the beginning of year t), the book-to-market ratio (book value of equity divided by market value of equity at time t), interest rate (the expected return on 5-year risk free government bonds), stock price volatility (the standard deviation of monthly returns on Oslo Stock Exchange) and net extraordinary items (total extraordinary items at time t divided by the market value of equity at the beginning of year t).

Define R 1 as the explanatory power from a regression that only includes the earnings variables, and R 2 as the explanatory power from a regression that only includes the control variables. The incremental value relevance of

**earnings, R 2, and the control variables, R CON, can then be defined as:**

EARN

Table 5 shows that explanatory power increases substantially in all regressions as control variables are added to the specifications. However, it is the incremental explanatory power of the earnings variables that is the measure of the accounting variables’ ability to explain stock returns. This incremental explanatory power is almost constant across different earnings aggregation levels for positive earnings. However, for the negative earnings sample, the incremental explanatory power of the earnings variables increases substantially as earnings are increasingly split into components. Incremental explanatory power increases from 11.16% to 11.71% as positive earnings are disaggregated, while the increase is from 0.90% to 7.93% when negative earnings are split into components. These findings support my hypothesis.