Specification problems with information content of earnings: revisions and rationality of expectations and self‐selection bias

This paper examines three possible specification problems with the research on information content of earnings disclosure. The first deals with the extent to which contemporaneous prediction errors are good surrogates for revisions of future earnings expectations and, hence, distributions of cash flows. This problem is elucidated by evaluating analysts' revisions of future earnings expectations as “relevant omitted variables.” The results show that the quality of such surrogation is high in the first quarter but low in the second. The second problem concerns the degree to which analysts' earnings forecasts are good surrogates for the market's own earnings expectations. Unbiasedness and orthogonality are the two properties examined. Although analysts' forecasts satisfy the unbiasedness property, the necessary condition of orthogonality is not satisfied. Hence, analysts' earnings forecasts are not Muthian rational expectations (i.e., they are not good surrogates for market forecasts). Consequently, the explanatory power of known empirical results is likely to be understated.

The third specification issue is the significance of the self‐selection bias resulting from endogenous partitioning of samples into, say, good‐ and bad‐news portfolios. The Heckman‐Lee method of correcting for this type of selection (truncation) bias is applied. The results show significant self‐selection bias in both quarters but more so in the first than in the second quarter. Although applying the correction for this sample did not alter the general inferences, it did alter the marginal contribution of each explanatory variable and the explanatory power of the models.

The results indicate that the information news about quarterly earnings is not homogeneous across different quarters in a fiscal period. The possibility that a “quarter effect” exists needs further investigation.