The Predictive Ability of Statistically-Based Cash-Flow Models: Working Paper Series--09-02

We assess the inter-temporal predictive ability of statistically-based, cash-flow prediction models by extending extant work on annual cash-flow prediction models. Our empirical results consistently underscore the superiority of quarterly cash-flow prediction models estimated on a time-series basis versus cross-sectional models. The superiority of relatively parsimonious, time-series models is consistent with the need to incorporate the firm-specific variability of parameters into expectations rather than restricting such parameters to be constant across firms and time when models are estimated cross-sectionally. Additionally, parsimonious models that employ aggregate earnings data are superior to more complex, disaggregated accrual models. The above results are similar regardless of whether models are estimated using undeflated or deflated variables. These results are particularly salient to researchers and users interested in generating accurate multi-step ahead cash-flow forecasts