Do Firm-Specific ERCs Help Explain Price Responses to Earnings News?

Prior studies consistently report higher explanatory power (R) for firm-specific returns-earnings models than for pooled models. The improvement in R by adding firm-specific coefficients to returns-earnings regressions is cited as evidence that price responses to earnings news vary with certain firm-specific characteristics. We argue that relying on R as a criterion to evaluate competing returns-earnings models could produce spurious results. Our empirical tests seek to determine whether the R improvement documented in earlier studies can be attributed to real economic characteristics that differ across firms. We begin by comparing Rs from actual-firm regressions to Rs from pseudo-firm regressions. Pseudo-firms are created from random draws of returns-earnings observations from pooled data. We find that actual-firm Rs do not exceed Rs for pseudo firms. Accordingly, the R advantage of firm-specific models over pooled models does not imply that time-series models capture cross-firm differences in economic characteristics. We also compare out-of-sample prediction errors between firm-specific and pooled models. Out-of-sample tests indicate that pooled models provide a better specification of the returns-earnings relation than firm-specific models. Accordingly, we conclude that crossfirm differences reported in time-series regressions are unique to the sample under examination – i.e., significant within-sample differences in slope coefficients lack external validity.