Loss Function Assumptions in Rational Expectations Tests on Financial Analysts' Earnings Forecasts

Abstract

Prior research concludes that financial analysts do not process public information efficiently in generating their earnings forecasts. The OLS regression-based tests used in prior studies assume implicitly that analysts face a quadratic loss function, or that analysts minimize their squared forecast errors. In contrast, we argue that analysts face a linear loss function, or that they minimize their absolute forecast errors. We conduct and compare rational expectations tests conditioned on these two alternative loss functions. While we replicate prior findings of inefficiency with OLS regressions, we find virtually no evidence of forecast inefficiency with Least Absolute Deviation regressions, where we explicitly assume a linear loss function.

Keywords: unbiased, economic significance, analyst rankings, conditional median, performance evaluation

JEL Classification: G10, G29, M41, D84