Predicting Earnings Using a Model Based on Cost Variability and Cost Stickiness

We evaluate the descriptive validity of the cost behavior model for profit analysis using Compustat data. For this purpose, we propose an earnings forecast model decomposing earnings into components that reflect (1) variability of costs with sales revenue and (2) stickiness in costs with sales declines. We evaluate the predictive ability of our model by benchmarking its performance in forecasting one-year-ahead returns on equity against that of two other time-series models based on line item information reported in the income statement and in the statement of cash flows. Specifically, we consider a model that disaggregates earnings into operating income and non-operating income components and another that disaggregates earnings into cash flows and accruals components. While all three models are less accurate than analysts' consensus forecasts that rely on a larger information set, we find that our model provides substantial improvement in forecast accuracy over the other two models that use only the line items in the financial statements. Finally, invoking the market efficiency assumption, we find that earnings forecast errors based on our model have greater relative information content than forecast errors based on the two alternative models based on financial statement information in explaining abnormal stock returns.