This dissertation questions the empirical finding that US equity returns have been predictable over the last 130 years, especially in the long run.; To this purpose, Chapter 2 examines the theory behind the use of dividend and earnings yields as proxies for expected returns, and shows that earnings yields have clear empirical advantages over dividend yields. In particular, given that prices exhibit long run excess volatility with respect to dividends, the use of the dividend yield as a predictive variable leads to a bias in the forecasting regressions, with the bias resulting in the Efficient Market Hypothesis (EMH) being unduly rejected. Taking account of these facts, simple test of stock market efficiency can be constructed. In the short run, the US equity market behaves as predicted by the EMH.; Chapter 3 develops a novel theory for mean reversion tests that is valid both under the EMH, and under the alternative that stock returns exhibit autocorrelation of unknown form. Applying these tests to US stock market data shows that there is some evidence for mean reversion at (medium-run) business cycle frequencies; however, evidence for long run mean reversion is weak at best. Conventional test results could have been highly misleading, suggesting that stock returns exhibit large short run variability and highly significant mean reversion in the long run.; In Chapter 4, a new theory is proposed for analyzing long run multivariate regressions. The method is computationally simpler and therefore more robust to small sample problems than existing methods. An efficient forecasting model is constructed and applied to US asset market data. It is shown that the long run component of the price earnings ratio, together with a mean reversion component, predicts up to 72% of the long run variation in S&P 500 returns, and up to 41% of the long run variation in US treasury yields. Most of this predictability seems to be due to a cycle of approximately 30 years that has repeated itself 2.5 times since the 1920's.
展开▼
