A Specification Analysis Of The Unexpected Earnings Response Regression Model PDF Download

Many studies have attempted to identify firm specific characteristics that influence the return-earnings relationship. The R-squared reported in these studies, however, are generally low and the earnings response coefficients are less than the theoretical value. Three factors have contributed to the low R-squared and the varying estimates of the earnings response coefficients: (1) violation of the OLS regression assumptions, particularly the linearity assumption, (2) omission of important (and correlated) variables, and (3) subjective specification of what variables (and interaction terms) to include in the earnings response regression.This study attempts to avoid these problems. Specifically, I examine all firm-specific characteristics that are known to influence the relationship between earnings and security return. This reduces the likelihood of an omitted correlated variable problem. More importantly, I use the recursive partitioning technique developed by Breiman et al. (1984), thus avoiding the subjectivity in specifying the earnings response regression model as well as the problems resulting from the distributional assumptions of the ordinary regression analysis.The recursive partitioning analysis indicates that five factors - size, market to book, earnings predictability, earnings persistence, and book value per share - affect the return-earnings relationship either individually and/or in interaction with each other and unexpected earnings. When included in the earnings response regression, these factors and their interactions explain a large percentage of the cross-sectional variations in return. Sensitivity tests are performed for alternate return windows and measure of earnings.