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

The primary focus of this study is not the over-time development in the predictive ability and value relevance of accounting measures. Therefore, I do not discuss possible reasons for the change. Note, however, that untabulated results show that the percentage share of firms that report negative earnings has increased over the sample period. In addition, more companies report impairment expenses in the second period than in the first. On the other hand, there seems to be a slight decrease in companies reporting extraordinary items on the income statement. As for the balance sheet, there has been a significant increase in companies reporting capitalised intangible assets.

recommended by Gu (2007). Scale-adjusted RMSE gives exactly the same results as the ones reported in table 4. In other words, predictive ability has decreased while value relevance has increased when scale-adjusted RMSE is applied as the measure of explanatory power.

6 Robustness Checks Several alternative tests are performed in order to test the robustness of step 1 and step 2 of this study. The robustness checks generally confirm the results from the main analysis. All tests are run for the positive and negative earnings sub-samples in addition to the total sample.

However, as all results are very similar to the main analysis, I present results only for the total sample.

6.1 Identical Samples Regression specification (2) demands fewer observations than specifications (1a) and (1c), which again demand fewer observations than specifications (1b) and (1d). As a result of this, the number of observations varies across the different specifications. The differences between the various regression analyses can theoretically be the result of different samples employed in the regressions. The first robustness check controls for this. In this test, only observations that are without missing values for any of the regression variables are used in every test. The results are presented in table 5.

Table description Table 5 describes the predictive ability and value relevance of earnings split into cash flow and accruals for a sample of Norwegian firms in the period 1992 to 2004. It summarises the regression coefficients (Coeff.), White-adjusted t-values (t-value), total explanatory power (adj. R2) and number of observations (n) for the total sample. The table presents the regression results when identical samples are used in all regressions, i.e., only observations with no missing values for any of the regression variables are

**used. Predictive ability (step 1) is analysed using regression specification (1a) to (1d), while value relevance (step 2) is analysed using regression specification (2):**

CFi,t is cash flow from operations for company i in year t, ACC is total accruals, EARN is earnings before extraordinary items and RET is stock return. ∆ denotes yearly change in the variables. The accounting variables are scaled by the market value of equity at 30 December in year t-1. Coefficients marked in boldface denote a statistical significance at a 5 % level, two sided test.

Table 6: Predictive Ability and Value Relevance of Cash Flow and Accruals - Market Values From March

Table description Table 6 describes the predictive ability and value relevance of earnings split into cash flow and accruals for a sample of Norwegian firms in the period 1992 to 2004. It summarises the regression coefficients (Coeff.), White-adjusted t-values (t-value), total explanatory power (adj. R2) and number of observations (n) for the total sample.

**Predictive ability (step 1) is analysed using regression specification (1a) to (1d), while value relevance (step 2) is analysed using regression specification (2):**

CFi,t is cash flow from operations for company i in year t, ACC is total accruals, EARN is earnings before extraordinary items and RET is stock return. ∆ denotes yearly change in the variables. Note the following changes from previous tables: RET is stock return from 31 March in year t until 31 March in year t+1. All CF, EARN and ACC data are scaled by the market value of equity per 31 March in year t. Coefficients marked in boldface denote a statistical significance at a 5 % level, two sided test.

predictor, while accruals are not. Both cash flow and accruals are related to future accounting earnings. The change variables are significant predictors of neither cash flow nor earnings.

However, all explanatory variables are significantly associated with stock return.

6.2 Stock Prices Measured in March The main analysis used stock prices measured in December each year. Several value relevance studies employ stock prices measured some time in year t+1, arguing that financial reports are not published on 31 December. Therefore, it takes some time for accounting information to become publicly known among the stock investors. As an alternative test, stock returns are measured from 31 March28 in year t to 31 March in year t+1. Note that this change does not only influence the stock return figures. All cash flow, accruals and earnings data are also changed since the variables now are deflated by the market value of equity on 31 March and not 30 December. Consequently, all regressions have to be re-run. The results are found in table 6.

Some of the regression parameters have changed from the main analysis. Accruals are now significantly negatively related to the mean of the next three cash flows. As before, cash flow is a significant cash flow predictor. Both cash flow and accruals are related to future earnings.

All explanatory variables except for the change in cash flow are significantly associated with stock return.

Note that the explanatory power is much lower in this value relevance regression. The accounting variables are able to explain very little of the stock return measured in March each Again, only actual trade prices are employed. Market data for the most illiquid stocks might be measured a few days prior to 31 March.

year. Timing is essential in value relevance studies. Aboody et al. (2002) suggest that stock markets are inefficient and react with a time delay to publicly available accounting information; they propose a way to overcome this inefficiency. Specifically, they multiply stock prices with the ratio of one plus the actual stock return to one plus the required rate of return, both measured in the future period τ, and claim that this procedure adjusts stock prices for predictable future price changes. They apply τ equal to 12, 24 and 36 months in their empirical analysis. In my study, the timeliness seems to have decreased as stock price measurement is delayed from December to March. This might be an indication that the stock market reacts quicker than Aboody et al. (2002) assume, and an indication that market inefficiency is not an issue in this kind of value relevance research. It might also be seen as evidence that the main analysis is more trustworthy than this robustness check.

6.3 Excess Return The main analysis employs raw returns as the measure of stock returns. Although there is no explicit standard for what return measure to use in value relevance research, some might argue that the market return must be controlled for in the regression specifications. For instance, a company may perform well even in years in which its stock has had a negative return, since all stock returns tend to be negative when the stock market plummets. As such, raw returns may understate the true value relevance of accounting information. Following Dechow (1994), all value relevance regressions are re-run using RET defined as stock return minus market wide return. Market wide returns are estimated from OSEBX29 – Oslo Stock Exchange Benchmark Index – per 30 December each year. Mean yearly market return is 15.7 % in the 1992-2004 period.

OSEBX is a value-weighted, investable index consisting of a representative selection of exchange listed companies at Oslo Stock Exchange. OB Total – Oslo Stock Exchange’s all shares index – is used to represent market wide returns for the period 1992-1995.

Table description Table 7 describes the value relevance of earnings split into cash flow and accruals for a sample of Norwegian firms in the period 1992 to 2004. It summarises the regression coefficients (Coeff.), White-adjusted t-values (tvalue), total explanatory power (adj. R2) and number of observations (n) for the total sample. Data are analysed

**using the following regression specification:**

(2) RETi,t = β 0 + β1CFi,t + β 2 ∆CFi,t + β 3 ACC i,t + β 4 ∆ACC i,t + ε i,t RETi,t is the excess stock return for company i in year t, CF is cash flow from operations and ACC is total accruals. ∆ denotes yearly change in the variables. The accounting variables are scaled by the market value of equity at 30 December in year t-1. Note the following change from previous tables: RET is in table 7 defined as excess return = stock return – market wide return. Market wide return is computed using OSEBX, a valueweighted, investable index consisting of a representative selection of exchange listed companies at Oslo Stock Exchange. OB Total – Oslo Stock Exchange’s all shares index – is used to represent market wide returns for the period 1992-1995. Coefficients marked in boldface denote a statistical significance at a 5 % level, two sided test.

The change in stock return definition does not influence the analysis of the variables’ ability to predict future cash flows and earnings. Therefore, only the value relevance regression is presented in table 7. Table 7 is practically identical to step 2 of table 3. Using excess returns instead of raw returns does not influence the study’s conclusion.

6.4 Other Robustness Checks Some alternative statistical procedures have been used in order to test the robustness of the

**conclusions (untabulated):**

• Newey-West standard deviations that control for possible autocorrelation in the data

before the study was conducted, a small number of observations may still be influential on the results. I have run robust regressions on the sample to test for the possible effect of outliers. Robust regression first performs an initial screening based on Cook’s distance 1 to eliminate gross outliers before calculating starting values and then performs Huber iterations followed by biweight iterations (StataCorp, 2005).

• Panel data techniques that apply generalised least squares have been performed.

All tests show that short term cash flow and earnings predictions are not equivalent to value relevance studies. Cash flow and accruals are significantly associated with current stock return. Current cash flow is consistently related to future cash flow and earnings, while accruals’ significance level remains dependent on whether cash flow or earnings are used as measures of future firm performance.

7 Concluding Remarks Step 1 of this study investigates the predictive ability of cash flow and accruals for short term firm performance as measured by future cash flow and earnings. The empirical findings suggest that cash flow is consistently related to both future cash flow and future earnings.

Accruals appear to be associated with future earnings but not with future cash flow. This can be seen as evidence against the FASB assertion that accruals make current earnings a better cash flow predictor than current cash flow. Step 2 of the study investigates the value relevance of cash flow and accruals. Both variables appear to be highly related to stock return.

The results of step 1 and 2 are, however, dependent of the sign of earnings. Consistent with Hayn’s (1995) assertion that negative earnings are less persistent than positive earnings, neither cash flow nor accruals are generally associated with future cash flow and earnings

the findings of step 2. I predict that if accruals and cash flows are related to short term future firm performance as measured by accounting earnings and cash flow, it is reasonable to expect that they are also significantly associated with current stock return. The findings are overall consistent with the prediction. However, the results strongly suggest that the prediction cannot be reversed. Accounting variables do not need to be associated with short term future cash flow or earnings, even if they are value relevant.

Much empirical accounting literature focuses on the predictive ability of accounting measures with respect to future cash flow and/or earnings. While some of the studies explicitly state that such prediction studies are regarded as substitutes for value relevance studies (e.g., Finger, 1994), this assumption is more implicitly observed in other prediction studies (e.g., Barth et al., 2001). As such, the prediction studies can be viewed as indirect value relevance studies. Kim and Kross’ (2005) study is one of few analyses that relates direct value relevance studies to indirect ones. Specifically, they compare the over time development in earnings’ ability as a short term cash flow predictor with the development in earnings’ value relevance.

They are surprised to find that earnings have become more related to one-year-ahead cash flows in a time period where earnings’ value relevance has decreased. My study extends the analysis of Kim and Kross (2005). Specifically, I study earnings’ ability to predict the mean of the three next cash flows, as well as one-year ahead earnings and the mean of the three next earnings. My findings are qualitatively identical to Kim and Kross (2005). The conclusion is that results from the cash flow and earnings predictions merely provide indications with respect to accounting variables’ value relevance. There is no one-to-one relationship between cash flow and accruals’ ability as cash flow and earnings predictors and their value relevance.

accounting variables’ relation with stock return.