Long-Horizon Mean-Reverting Stock Prices Revisited
重新检验长期股票收益，发现先前研究夸大了均值回归的证据，主要源于普通最小二乘法对经济大萧条和二战时期数据的过度加权，以及依赖渐近统计和只关注最负的均值回归估计。使用广义最小二乘随机化检验，1926至1987年及1871至1987年的数据均无法拒绝随机游走假说。
This paper reexamines long-horizon stock returns and finds that previous work overstates the evidence of mean reversion. The overstatement is largely due to the implicit weighting of ordinary least-squares tests, which place more weight on the Depression and World War II observations, which have both large error variances and stronger mean-reverting tendencies. Additionally, the reliance on asymptotic statistics and the improper focus on only the most negative estimates of mean reversion contribute to the overstatement. Using generalized least-squares randomization tests on the 1926 to 1987 period, the random walk cannot be rejected for value- or equally-weighted real returns at any of 10 return horizons or by joint tests over all 10 horizons simultaneously. Additionally, the random walk cannot be rejected for the extended 1871 to 1987 period.