| The exploration into stock return predictability has its origins far back to the Dow Theory and the Random Walk Theory. The first suggests existence of short persistence but reversion in long horizons, while the second suggests unpredictability of returns. This unpredictability is rationalized in the fair-game model as the result of market equilibrium of demand and supply. Regarding the pattern of autocorrelation of stock returns, there are two different interpretations: (1) Financial market inefficiency induces large temporary irrational swings of stock prices, thus creating slowly decaying (mean-reverting) components; (2) Time-varying expected returns create transitory components of stock prices, which doesn't necessarily violate market rationality. Despite the important implication of mean reversion, most empirical studies have not been so successful in detecting significant evidence of it.;Moreover, duration analysis of the stock market cycles show evidence of negative duration dependence (positive autocorrelation) in short-horizon durations, but positive duration dependence (negative autocorrelation) in longer-horizon durations. Bootstrap results are all consistent with, and confirm the actual regression results. Based on all the test results, we conclude that mean reversion of stock prices is not any statistical illusion but a reality in the recent US financial markets, which appears to be closely related to the self-regulating mechanism of the recent US economy.;The recent years are reported to be characterized as steady expansion and a self-regulating US economy. Considering stock market indexes are one of the most important indicators of national economic performance, a similar pattern might appear in stock indexes. This is summarized by the linked testable hypothesis that a well-functioning stabilization mechanism of the recent US economy should appear in stock markets as the self-regulating or mean-reverting behavior. The estimation results based on the Newey-West statistics show evidence of substantial negative autocorrelations in long-horizon returns during the recent 10 year period, which appears stronger for big stocks. The results of multivariate test show that a portion of mean reversion can be decomposed by the positive cointegration between prices and dividends. In addition, they suggest that mean reversion can be explained as changes in portfolio risk and subsequent compensation in expected returns. |
