«Leif Atle Beisland University of Agder Dissertation submitted to the Department of Accounting, Auditing and Law at the Norwegian School of Economics ...»
Note that if ignoring discounting, step 1 and step 2 will measure exactly the same thing if indefinite cash flow and earnings were included as dependent variables in step 1. However, a main purpose of this study is to investigate whether or not relatively short term cash flow and/or earnings predictions can act as proxies for value relevance analyses. Since the focus in Barth et al. (2005) state that median equity prediction errors are smallest when earnings are disaggregated into cash flow and total accruals.
these attributes are not discounted20.
The econometric models use cash flow, accruals, and their associated changes as explanatory variables. Easton and Harris (1991) show that, depending on the valuation model employed, stock return can be seen as a function of both earnings and the change in earnings. The residual income model can easily illustrate this important point. This model says that equity value is a function of book value of equity and accounting earnings (see Ohlson, 1995).
Change in equity value, which is equivalent to stock return, is a function of the change in book value of equity and the change in earnings. The change in book value of equity equals earnings when no dividends are paid. Hence, change in equity value (stock return) is a function of both earnings and change in earnings21.
Prior research shows that the value relevance of earnings may be a non-linear function of the earnings level (for instance Freeman & Tse, 1992). Specifically, research widely documents that value relevance is dependent on the sign of earnings (Basu, 1997; Joos & Plesko, 2005).
Negative earnings are hardly related to current stock returns at all. This finding is often attributed to the liquidation option that investors hold (see, e.g., Hayn, 1995). Investors will liquidate incorrigible loss firms rather than suffer from indefinite losses. If loss firms survive, it must be because investors expect the losses to be temporary. The lack of persistence of negative earnings may not only affect the value relevance regressions. Obviously, if negative Obviously, if direct and indirect studies of value relevance cannot be reconciled, one of the reasons may be the lack of present value calculations in cash flow/earnings prediction studies. However, since discount rates are unobservable in the stock market, such calculations may be rather noisy: “We acknowledge that estimating discount rates is susceptible to measurement error” (Subramanyam & Venkatchalam, 2007, p. 464).
Easton and Harris (1991) perform an empirical study where both earnings and change in earnings are used as explanatory variables for return. They find that earnings levels are significantly associated with stock return. As for earnings changes, the regression coefficient is statistically significant in slightly less than half of the years analysed. Lev and Zarowin (1999) employ the same methodology for cash flows, but they report only the combined slope coefficients for the two cash flow variables.
the sign of earnings. Ball and Shivakumar (2006) claim that linear specifications may understate the ability of current earnings to predict future cash flows. As such, the predictive ability and value relevance of cash flow and accruals are analysed not only for a pooled sample but also for positive and negative earnings sub-samples. This will assure that the sign effect has been controlled for when the study’s conclusions are stated.
4 Data The sample consists of firms listed on the Oslo Stock Exchange. To ensure consistency with sample-selection criteria used in prior studies, the sample excludes financial services firms.
All accounting data are obtained from the Oslo Stock Exchange accounting database for exchange-listed companies. Stock price data are collected from the Norwegian School of Economics and Business Administration’s Stock Market Database. All stock prices are adjusted for dividends, splits, etc. Stock values and –returns are measured on 30 December each year22.
Observations are from 1992-2004. In 1992, the 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 time23. In 2005, European law required Norwegian quoted companies to report consolidated statements according to International Financial Reporting Standards (IFRS). Since the introduction of IFRS may have influenced both the structural relationship between stock return and earnings numbers, as well as earnings numbers’ ability to predict themselves, I do not include the IFRS observations in 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 also that older data might have suffered from poor liquidity at Oslo Stock Exchange (OSE). While the value weighted percentage turnover at OSE was at approximately the same level in 2002 as in 1992, the turnover was almost twenty times higher in 1992 than in 1982 (Næs, Skjeltorp, & Ødegaard, 2008, p. 4)!
belonging to the upper or lower percentile of RET, CF, ∆CF, ACC and ∆ACC are deleted to avoid extreme observations with unreasonably large influence on the regression results. Due to a large degree of overlap among extreme observations, the actual number of observations deleted is 77 (4.9 %), far less than the theoretical maximum of 10 %. The final sample size is 1587 observations. The maximum number of available observations is applied for all main regressions. However, as a robustness check, I repeat the analyses using a constant sample.
Table 1 shows the main descriptive statistics of the variables employed in the analysis. The distributional characteristics are found in panel A. Mean cash flow from operations is equal to 12.9 % of the beginning market value of equity. Mean accruals are -11.7 %, while average earnings are only 0.8 % of market value of equity in this sample. Actually, untabulated results show that mean earnings are negative in 7 out of 13 years. The standard deviations for cash flow and accruals are quite high, while the variation in net income is lower. Accruals do somewhat level out the cash flow variations. As for the change variables, the mean change in cash flow and accruals seem moderate. However, the standard deviations are higher than for the original variables. Even though the change in cash flow and accruals fluctuates around zero on average, the dispersion is definitely substantial. As expected, the volatility of an average is lower than that of an individual observation. Hence, the means of the next three cash flows and accounting earnings have substantially lower standard deviations than the yearly observations. Note that the mean stock return is very high for the period in question.
Nevertheless, the risk, as measured by the annual standard deviation, is considerable.
All CF, EARN and ACC data 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.
accruals seem to be highly negatively correlated. This is further evidence that accruals contribute to level out the cash flow variations. In addition, the change in cash flow is significantly negatively correlated with the change in accruals. Earnings are significantly correlated with cash flow, but they have a lower correlation with accruals. The measures of future firm performance are generally correlated with today’s cash flow and accruals but not with the first difference of these variables. All measures of future firm performance are significantly correlated with each other. Note that none of the variables have a particularly high correlation with stock return. Specifically, accruals seem to have a low association with stock return.
5 Main Empirical Results
5.1 Step One: Cash Flow and Earnings Predictions The first step of the study analyses the predictive ability of current cash flow and accruals with respect to future firm performance as measured by cash flow and accounting earnings.
Table 2 summarises the empirical findings from these regressions. In the total sample, cash flow seems to be a significant predictor24 of future cash flows. As for next year’s cash flow, both the level and the first difference of current cash flow turn out to be significant predictors.
However, accruals seem to be unrelated to future cash flow. In fact, this conclusion holds with respect to next year’s cash flow as well as the mean of the next three cash flows. The mean of the next three cash flows appears to be easier to forecast than next year’s cash flow, as the explanatory power is 2.5 times higher in regression (1b) than in (1a)25. Overall, these results indicate that there is positive auto correlation for cash flows; a high cash flow in one year is The presented t-values are computed using White-adjusted standard deviations. 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.
Note that the sample size differs between the regressions. As a robustness check, all regressions are re-run on an identical sample, compare section 6.1.
research (Barth et al., 2001; Dechow et al., 1998). Even though a high cash flow in the normal case is followed by another high cash flow, a high increase in cash flow seems to have a negative impact on future cash flows, at least next year’s cash flow. This finding indicates that the cash flows to a certain extent mean revert. The indication that companies performing badly for some time tend to perform better in the future, and vice versa, is a phenomenon frequently discussed in capital market-based accounting research (see for instance Ball & Brown, 1968; Basu, 1997; Hayn, 1995; Sloan, 1996).
Table 2: Step 1 - Predictive Ability of Cash Flow and Accruals
Table description Table 2 describes the predictive ability 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 as well as for the
where CFi,t is cash flow from operations for company i in year t, ACC is total accruals and EARN is earnings before extraordinary items. ∆ 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.
Contrary to Barth et al. (2001) and Dechow et al. (1998), table 2 suggests that there is no significant relation between future cash flows and current accruals in the total sample. Neither accruals nor the change in accruals shows significant coefficients in the regressions. It does not matter whether the dependent variable is next year’s cash flow or the average of the next three cash flows. The explanatory power of the two specifications is 12 % and 29 %, respectively. When only cash flow and its first difference are used as explanatory variables, the adjusted R 2 is respectively 12 % and 28 % (untabulated). These results contradict the assertion that accruals make current earnings a better predictor of future cash flows than current cash flow. When the sample is split according to the sign of earnings, accruals are significantly related to next year’s cash flow for the positive earnings sample. Accruals remain unable to predict the mean of the next three cash flows. Consistent with prior research (for instance Hayn, 1995), today’s cash flow and accruals have a low association with future cash flow when earnings are negative. Actually, none of the regression coefficients are significant when the mean of the three next cash flows is analysed. Explanatory power is also
investigated. Note that cash flow may be both positive and negative when the sample is split – the split is governed by the sign of the earnings.
Table 2 also displays the findings from regressions of future earnings on today’s cash flow and accruals. The results for the total sample clearly indicate that accruals are a relevant earnings predictor. In fact, both cash flow and accruals are significantly associated with future firm performance when cash flow is replaced by earnings as the dependent variable. Though cash flows are positively related to future earnings, large levels of accruals are typically associated with lower future earnings (note that total accruals typically are negative). Still, the change in accruals is statistically unrelated to future earnings. This is also the case for the change in cash flow. The explanatory power is the same in regression (1c) and (1d). In contrast to the cash flow regressions, it does not appear to be easier to forecast the mean of the next three earnings than next year’s earnings. This finding may be attributed to the higher variation in cash flow than in earnings (see table 1). Accruals contribute to levelling out earnings but not cash flow. Note that untabulated results show that the explanatory power of the earnings regressions falls dramatically if either cash flow or accruals are omitted as an explanatory variable. For instance, the adjusted R 2 is only 2 % when next year’s earnings are regressed on either cash flow or accruals.
As with cash flow predictions, earnings predictability for future earnings is sign dependent (see table 2). All explanatory variables are significant when earnings are positive and next year’s earnings are to be forecasted. When the mean of the next three earnings is predicted, only the level of cash flow and accruals is significant. In the case of negative earnings, the mean of the next three earnings seems practically unpredictable. Only the level of accruals is
power from the earnings prediction regressions is lower when earnings are negative than when they are positive, but the explanatory power is less sign dependent than in the cash flow predictions.