Implications of scale invariance in EPS forecast errors

Errors associated with analysts' forecasts of earnings per share (EPS) would vary naturally with scale, but managers reverse that variation by differentially suppressing forecast error magnitudes, with high price firms compressing to a greater extent. As a result, little variation with scale is observed not only for EPS forecast error magnitudes, but also for volatility of core and reported EPS, and dispersion of EPS forecasts. We investigate two implications of this behavior. First, the intuitive practice of deflating those variables by scale creates a strong negative correlation with scale. Using deflated variables as dependent (independent) variables in analyses that include independent (dependent) variables that happen to be correlated with scale might bias coefficient estimates. Second, price responses to EPS surprise (or ERC) for the highly compressed surprises of high price shares can exceed 90, which are much greater than ERCs for low price firms as well as levels expected in prior research. This strong positive relation between ERC and scale might bias studies investigating cross-sectional variation in ERC if that variation is related to scale. More generally, pooling subsamples with very different characteristics, such as low and high price firms, potentially obfuscates important differences.