Cross-sectional estimation: Fama and French (2000) introduce a crosssectional
estimation approach to the earnings forecasting literature to uncover
the time-series properties of earnings.They argue that time-series estimation
lacks power because there are only a few time-series observations of annual
earnings available for most firms.In addition, use of a long time series
introduces survivor bias.The survivor bias implies more observations of
positive earnings changes following positive changes than expected by chance,
for reasons discussed above.This offsets the underlying negative time-series
correlation in earnings changes.The effect of survivor bias, together with low power (i.e., large standard errors) of time-series estimation, favors the
conclusion of a random walk in annual earnings.