In this study, we investigate why earnings underperform relative to cash flows in explaining future cash flows. Using analytical and empirical analyses of the components of R2 values, we demonstrate an ‘opposing covariances effect’ in which covariances of opposite sign are summed and thus offset to reduce R2 values in cash flow prediction models. Counterintuitively, this opposing covariances effect causes cash flow prediction tests using aggregate earnings to exhibit higher R2 values when accruals covary less with future cash flows, all else equal. Understanding this opposing covariances effect reconciles certain empirical discrepancies and changes inferences on multiple fronts. First, we challenge the inference based on R2 values that earnings fail to accomplish central tenets of financial reporting (Nallareddy et al., 2020). Second, we draw similar inferences when examining earnings' ability to explain both contemporaneous returns and future cash flows (Dechow, 1994). Third, we show that the increase over time in earnings’ ability to predict future cash flows is driven by attributes of accruals, not cash flows (Kim and Kross, 2005; Nallareddy et al., 2020). Fourth, we show that the improved predictive ability of the adjusted earnings measures proposed by Ball and Nikolaev (2022) occurs because their adjusted accruals covary less with future cash flows. Fifth, while we agree with Barth et al. (2001) that aggregating earnings masks information, we show that the opposing covariances effect is an important driver of this masking, not only the equal weighting of earnings components. Overall, our findings suggest that future research examining earnings as a summary measure should not rely on R2 values alone to infer the extent to which accrual-based measures convey relevant information or achieve the intent of SFAC No. 1.

zcarol2@ceibs.edu