Business Model Change and Cash Flow Forecast Accuracy

This study develops a text-based measure of business model change (BMC) derived from year-over-year semantic shifts in firms’ Item 1 disclosures and examines its implications for cash flow forecasting. Using a large U.S. panel with underlying data spanning 1996 to 2023 with over 55,000 firm years, the analysis relates BMC to the accuracy of three mechanical models, that is bottom-line earnings (M1), operating cash flow (M2), and operating earnings (M3), and to analyst forecasts. Forecast-comparison tests reveal a clear ordering: operating-based models (M2/M3) outperform M1 at all horizons; analysts hold a small one-year edge during the transition period, but from year two onward the operating-based models (M2/M3) are more accurate. Forecast errors increase (quasi-)monotonically across BMC deciles, and in cross-sectional regressions with rich controls, higher BMC predicts larger absolute errors, more optimistic signed errors, and greater analyst dispersion. Adding BMC as a predictor significantly improves out-of-sample performance: Clark–West tests indicate statistically significant reductions in forecast error at the one- and two-year horizons, with attenuating significance at the three-year horizon. Moreover, quantile regressions reveal a non-linear risk profile: the marginal effect of BMC on analyst forecast error quadruples as one moves from the median (Q50) to the tail (Q90), identifying BMC as a leading indicator of extreme valuation failures. Overall, BMC is a timely, interpretable signal of changing fundamentals that both raises forecast difficulty and adds incremental predictive content when incorporated into standard models.