<p>The prediction of cash flows from operations (CFO) is crucial for many applications in accounting and finance. This paper studies statistically-based cash flow forecasting. Traditionally, the literature on CFO forecasting has focused on either cross-sectional or time-series estimation methods in isolation. The cross-sectional regression-based approach has the advantage of minimal data requirements that serves to maximize sample size (n is generally large in practice). Meanwhile, the time-series-based predictive model accommodates firm-specific variability in beta but demands a sufficiently long time series of data. To address these limitations, we propose a novel predictive model leveraging the local learning approach. This method integrates time-series data while allowing beta to vary with firm size, adhering to the principles of local learning prediction. Our model contributes to the CFO forecasting literature by combining the strengths of cross-sectional estimation (using data from firms of similar size) with time-series regression (capturing firm-specific beta variability). Empirical results demonstrate that incorporating features from both approaches enhances CFO prediction accuracy when applied concurrently in our proposed framework.</p>