On the Predictability of Corporate Earnings Per Share Behavior

RECENT RESEARCH ON THE predictability of corporate annual income numbers has indicated that in general such series are best described as essentially random processes. These results have been confirmed by many studies, Albrecht et al [1], Ball and Watts [2], Brealey [4], Lintner and Glauber [15], Watts and Leftwich [20], and have been widely cited in financial research (see Ederington [9]). The conclusions of these studies have major implications for financial theories which rely on assumptions of income predictability. Theories of corporate valuation that rely on growth and historical predictions thereof, such as those of Gordon [11] and Lintner [14], are based on questionable premises if such growth cannot be predicted. Notions of multi-period income-smoothing, which depend on predictable net income series, such as those of Barnea et al [3], and Gordon [12], are also likely to be incorrect in such an environment. These are non-trivial consequences and the findings have received considerable attention for this reason. One limitation of the previous research is that the tests of predictability employed rely completely on extrapolatory or time-series models of annual EPS behavior (see all references above). The present research investigates the proposition that income numbers may have predictable properties when considered in their economic environment. To test this proposition, this study employs forecasting models utilizing economic lead-indicators to examine the predictability of annual EPS behavior. The methodology involves comparisons of the accuracy of one-step-ahead forecasts of individual firms' EPS numbers generated by alternative lead-indicator and extrapolatory models. The results indicate the existence of predictable (dependent) relationships between some economic lead-indicators and EPS numbers.