Abstract
Theoretically the expected return on any financial asset need not equal the average of the actual return. In this paper, the expected equity premium is estimated based on two fundamentals: the Gordon dividend model with constant growth, and duration analysis. The result is that the ex ante, or expected, equity premium is around 3.24%, with a standard error between 0.30% and 0.87%. Taking 0.87% as the standard error, the 95% confidence interval is between 1.53% and 4.95%. These figures show clearly that the actual equity premium is much higher than the expected one. The reason for that is due to unpredictable changes in interest rates, and other growth rates.
Acknowledgments
I am grateful to an anonymous referee and to his advice to rewrite and shorten the paper in order to stress the author's contribution and to cover quickly the literature on the subject because it is widely known.
Notes
1 The growth rate g could be the average growth rate in earnings or the average capital gain on the security.
2 The adjusted R-Square is 0.0445 with a total number of observations of 716. Serial correlation of the residuals is rejected. First-order serial correlation is tested by the Durbin–Watson statistic, which has a value of 2.064, close to 2. Higher-order serial correlation is tested by the Breusch/Godfrey statistics, and by the Ljung–Box Q-statistics, both for six lags. The lowest probability for these two tests is 0.343, which is larger than the cut-off probability of 0.05, the usual type I error, and therefore higher-order serial correlation is rejected up to six lags. The CUSUM test has a probability of 0.917, which is much higher than 0.05, the usual cut-off probability, and hence stability of the parameters is assured. The TSP® RESET test has a probability of 0.498 rejecting misspecification, the cut-off probability being 0.05. The TSP® heteroscedasticity test has a probability of 0.716 which rejects heteroscedasticity with high confidence, the cut-off probability being 0.05 also. The only poor econometric diagnostic is the non-normality of the residuals. However OLS is usually robust to limited departure from normality of the residuals.