Mean Reversion in Stock Prices? A Reappraisal of the Empirical Evidence
Myung Jig Kim, Charles R. Nelson, Richard Startz
NBER Working Paper No. 2795 (Also Reprint No. r1589)
Recent research based on variance ratios and multiperiod-return autocorrelations concludes that the stock market exhibits mean reversion in the sense that a return in excess of the average tends to be followed by partially offsetting returns in the opposite direction. Dividing history into pre-1926, 1926-46, and post-1946 subperiods, we find that the mean-reversion phenomenon is a feature of the 1926-46 period, but not of the post-1946 period which instead exhibits persistence of returns. Evidence for pre-1926 data is mixed. The statistical significance of test statistics is assessed by estimating their distribution using stratified randomization. Autocorrelations of multiperiod returns imply a forecast of future returns, which is presented for post-war three-year returns using 1926-46, full sample, and sequentially updated coefficient estimates. The correlation between actual and forecasted returns is negative in each case. We conclude that evidence of mean reversion in U.S. stock returns is substantially weaker than reported in the recent literature. If mean-reversion continues to be a feature of the stock market, then the experience of the past forty years has been an aberration.
Published: Review of Economic Studies, Vol. 58, No. 195, pp. 515-528, (May 1991).