The Role of Accruals in the Prediction of Future Earnings
Journal of Accounting and Finance
Although past accruals can explain current earnings, whether current accruals can predict future earnings has not been examined. I introduce three regression-based prediction models to predict future earnings: using only cash flows only, both cash flows and accruals, and earnings, to determine whether current accruals contribute to the prediction of future earnings. I find a cash flows and accruals model is more precise than a cash flows-only model, but the earnings model is the most precise.
... the most precise. This result suggests that current earnings are a more accurate predictor of future earnings, and that accruals' persistence is less important in prediction of future earnings. I also document that predictions based on these models are more precise than time series-based predictions, but not as precise as analysts' forecasts. Additionally, I investigate how current earnings management negatively affects prediction of future earnings and hypothesize that current accrual management and real earnings management are positively associated with the prediction errors of the three models. The result shows that only the measure of real earnings management is positively associated with prediction error. have different persistence; specifically, accruals are less persistent. Francis and Smith (2005) argue that the accruals used in research have implications in both current and non-current transactions and that the portion from non-current periods leads to lower persistence. In other words, current accruals, defined as earnings less cash flows, should have different (lower) persistence compared to current cash flows. This leads to the conjecture that earnings alone are not a better predictor than cash flows and accruals combined. Barth et al. (2001) argue that disaggregated accrual components can have a different impact on future cash flows and that earnings can mask the impact. Thus, the prior literature suggests that predictions based on cash flows and accruals separately (or cash flows and the components of accruals, Components model, hereafter) would also be more accurate predictors of future earnings in an out-ofsample setting. In contrast, in an out-of-sample setting, if a firm engages in earnings smoothing, regardless of earnings components, such as cash flows, accruals or components of accruals, this would produce a stable earnings trend. Therefore, prediction based on earnings alone can be a more accurate predictor in this context. Prediction based on cash flows alone ignores accruals and their persistence completely. Therefore, its accuracy should be lower, in theory. However, if accruals are only noise in the earnings, or are managed opportunistically, then cash flows-based prediction can be more accurate. If earnings management through accrual management is widespread, then accruals, on average, will be ineffective for forecasting. Next, I examine the association of earnings management in the current period and errors in predicting future earnings based on these models. The accounting literature, including work from Cohen et al. (2008) and Jones (1991), provides two earnings management mechanisms: accruals management and real earnings management. If current earnings are managed opportunistically by managing accruals or cash flows, then the future earnings predictions of these models would suffer higher prediction errors. For example, under opportunistic accruals management in the current period, accruals tend to reverse and can be less effective in forecasting future earnings. This will be revealed in the association of accruals management proxy in the current period and prediction error. A positively significant association between current accruals management measures and prediction error provides direct evidence implying that accruals management interferes in accurate prediction of future earnings based on financial statements. Real earnings management can also harm predictions of future earnings. Real earnings management, which can be achieved by changing a firm's operations to obtain favorable financial reporting, affects cash flows. By adjusting the timing of investments, Research &Development expenditures, and Selling, General &Administration expenses in current period, a firm's cash flows may increase, thus increasing earnings. However, this will obscure the true relationship between current cash flows and future earnings, thereby increasing the prediction error of the forecast when using cash flows as a predictor. Using the well-established proxy for accruals management (discretionary accruals, Jones 1991) and real earnings management (RMproxy, Cohen et al. 2008), I test whether these proxies in the current period are positively associated with error in predicting future earnings. The result shows that accruals have predictive power beyond cash flows in out-of-sample prediction. The CFO & ACC model's prediction is more accurate than that of the CFO-only model in predicting future earnings. The Earnings model, however, is more accurate than the CFO & ACC model. The Components model is the least accurate in predicting future earnings, though the in-sample R-squared of this model is highest. Thus, there is evidence that current accruals contribute to the prediction of future earnings beyond cash flows; however, current earnings are the most accurate predictor of future earnings. I also document following characteristics that have not been previously described in the literature. First, the four regression-based predictions described above outperform the random walk and seasonal random walk method by an average of 30%. Most time series earnings forecasts in the literature tend to focus on a specific industry or group of firms (see Bradshaw et al. 2012 for a summary). Using a relatively larger sample that represents overall populations, the CFO-only, CFO & ACC, Earnings and Components models are shown to outperform time series earnings prediction. This shows that regressionbased prediction can be more effective in predicting future earnings. 12. Analysts' forecast error (AFE) is the absolute value of analysts' forecast error, based on the first consensus forecast for quarter t+1, made after the earnings announcement date of quarter t-1, |Actual EPS t+1 -FORECAST FIRST, t+1 |,