The Explanatory Power of Earnings for Stock Returns.
通过面板回归改进收益与回报关系的模型设定，发现允许参数跨截面和时间变化、纳入当期收益率以及区分收益持续性成分，可将调整R方从0.10提升至0.38，反驳了Lev(1989)关于收益解释力过低的结论。
Abstract In a thorough review of market-based research on the information content of accounting earnings, Lev (1989) concludes that the explanatory value of earnings for stock returns, and therefore the usefulness of earnings disclosures, tends to be embarrassingly low. A number of nonmutually exclusive explanations have been advanced for these disappointing results, including: (1) poor specification of the estimating equation, such as a failure to allow for cross-sectional variation in the regression parameters; (2) inappropriate choice of the assumed proxy for expected earnings; (3) the availability of more timely sources of the value-relevant information in earnings statements (Beaver et al. 1980); and (4) poor informational properties (quality) of reported earnings because of biases induced by accounting measurement practices or creative "abuses" of the earnings measurement process. Lev (1989) speculated that the last of these explanations was the most likely cause of the poor statistical performance consistently found in returns-earnings research. In contrast, the present study shows that a considerable improvement in statistical performance can be achieved by working with a more general specification of the returns-earnings relation. Lev's article has resulted in serious questioning of the contribution of market-based research, but we believe that the present study provides grounds for a more positive assessment. We use a panel regression approach to examine the association between annual stock price returns and reported earnings figures of industrial companies in the United Kingdom. We combine several recent advances in market-based accounting research design to produce a specification of the relation between earnings and price changes that subsumes the following key features: 1. Contemporaneous earnings yield is included in addition to the deflated first difference in earnings that is normally included in models of the returns-earnings relation. 2. Regression parameters are allowed to vary both cross-sectionally and over time. 3. Parameter values are allowed to vary across components of earnings to accommodate differences in the degree of persistence; in particular, we model the explanatory power resulting from attempts by accountants to distinguish extraordinary and exceptional items from the other components of earnings. We introduce these features in a general model in a way that allows us to assess the incremental explanatory power of each individually as well as the joint effects of two or more combined. Each methodological improvement contributes significantly to our ability to explain security price changes, and we show that the best fit is achieved by incorporating all three features in a single general model. In moving from the standard model, which regresses a measure of abnormal returns on earnings changes, to the most general model, the adjusted A-squared increases from 0.10 to 0.38.