Time series versus cross-sectionally derived predictions of future cash flows

Abstract

Abstract We refine the analysis of annual cash-flow prediction models originally developed and tested by Dechow et al. (1998), Barth et al. (2001) and Kim and Kross (2005) using cash flow from operations data reported in accordance with FASB Standard No. 95 for a constant sample of 1111 firms. We estimated annual cash-flow prediction models both cross-sectionally and on a time-series basis to assess whether restricting firm-specific parameter estimation in the cross-sectional approach adversely affects predictive performance. Predictive ability is assessed via “out-of-sample” forecasts in an inter-temporal holdout period (2001–2005) not used in model estimation. We provide new evidence that significantly greater enhancement to predictive performance is obtained when cash-flow prediction models are estimated on a time-series basis versus cross-sectionally. These inferences are robust across one-year ahead cash-flow predictions or one-thru-five-year ahead predictions. We find that the relative accuracy of cash-flow predictions is unaffected by whether the aforementioned prediction models employ cash flows or net earnings as independent variables. Finally, we also provide evidence that the predictive ability of cash flows is highly sensitive to firm size. That is, relatively larger firms provide significantly more accurate cash-flow predictions than those of smaller firms across cash-flow prediction models.