| This study examines whether the accuracy of earnings components (accruals, AC and cash flows from operations, CF) in predicting future earnings components, instead of total earnings, improves earnings prediction and explains earnings persistence and the earnings-returns relationship. This study focuses on the combined-earnings-components prediction model: AC and CF are used as joint predictors to predict each earnings component separately and predicted values of AC and CF are summed to derive the predicted total earnings. This study compares the predictive accuracy among three earnings prediction models: the autoregressive model, the earnings-component-based prediction model and the combined-earnings-components prediction model. The results show that the earnings-component-based prediction model has the strongest ability to predict future earnings and the combined-earnings-components prediction model is the least accurate. This result indicates that AC and CF are correlated and the accuracy of earnings prediction is reduced if the correlation between earnings components is ignored. Also, this study documents that forecast errors obtained from the earnings-component-based prediction model are due in part to the inability of current earnings components to predict the separate future components. However, a more accurate prediction of future earnings components is not consistent with the higher earnings persistence and the stronger earnings-returns relationship. Similar to the Sloan's study (1996), this study attributes this discrepancy to the market failing to fully reflect the predictive ability of accrual and cash flow components of current earnings. Alternatively, the market may react more favorably to unexpected positive changes on earnings components. |
