I synthesize and discuss academic research on financial statement analysis and earnings forecasting. I begin by discussing analytical and empirical evidence that shows that earnings, not dividends or free cash flows, are the payoffs that investors forecast when estimating value. This result is fundamental and it provides clear motivation for studying earnings forecasting and the role that historical accounting numbers play in the earnings-forecasting process. I then provide a detailed discussion of the research design choices that are made when developing and evaluating an earnings-forecasting approach. I describe the tradeoffs involved when making these choices and I review the extant empirical literature. An overarching theme of this discussion is that there are substantial research opportunities.For example:
The random-walk model performs too well on a relative basis. It is inconsistent with standard economic assumptions, accounting practice and the way financial statement analysis is practiced and taught. Nonetheless, it tends to be as accurate and sometimes more accurate than other extant approaches.
Panel-data approaches that use a mix of cross-sectional and time-series data are very flexible in terms of the: (1) choice of earnings metric to predict; (2) choice of predictors; (3) choice of estimator; and (4) choice of estimation sample. At present, these approaches have not been used to their full potential.
There is insufficient evidence regarding how to identify peers and the role that peer analysis plays in the forecasting process.
There is insufficient evidence regarding approaches for forecasting the higher moments of future earnings, how to evaluate these forecasts and their role in determining value. Moreover, the role that accounting measurement plays in the determination of the higher moments of earnings and how accounting-measurement rules affect the usefulness of historical accounting numbers for predicting the higher moments of future earnings is not well understood.
