Loss function assumptions in rational expectations tests on financial analysts' earnings forecasts

Prior research concludes that financial analysts do not process public information efficiently in generating their earnings forecasts. The ordinary least squares (OLS) regression-based tests used in prior studies assume implicitly that analysts face a quadratic loss function. In contrast, we argue that analysts likely face a linear loss function, and hence, try to minimize their absolute forecast errors. We conduct and compare rational expectations tests using these two alternative loss functions. We reproduce most prior findings of forecast inefficiency with OLS regressions, but find virtually no evidence of forecast inefficiency with least absolute deviation regressions, where we explicitly assume a linear loss function. (C) 2004 Elsevier B.V. All rights reserved.