Stock Return Predictability with a Near-Random Walk Model: Evidence and Implications

• Activity Number: 83
• Type: Contributed
• Date/Time: Sunday, August 4, 2013 : 4:00 PM to 5:50 PM
• Sponsor: Business and Economic Statistics Section
• Author(s): Staffan Fredricsson*+
• Keywords: stock return ; prediction ; random walk ; unit root ; mean reversion

Abstract

Can predictions of real stock returns be improved beyond those offered by the Random Walk model? After more than a quarter-century, the answer to that question is still uncertain. In this paper, I use an AR(1) model with linear trend and a US stock index with annual data. This is an extension of the strict random walk model, since roots to the characteristic equation with value less than and including unity are supported. The well-known bias problem encountered when estimating the autoregressive parameter in the near unit root environment is addressed by using a median-unbiased method. Predictions for the near-random walk and strictly-random walk models are compared, and the results are found to favor the near-random walk alternative. The statistical significance of the differences, based on Monte Carlo simulations, are reported. The results not only confirm earlier findings that stock prices may be mean-reverting, but also offer a prediction method resulting in reduced variability from expected returns, and therefore a reduced risk. This should be of particular interest to those concerned with the investment of assets for retirement and other longer-term purposes.