Abstract: | A procedure that explicitly estimates the form of beta nonstationary followed by a security is examined. Beta estimates and forecasts generated from this process are compared with estimates and forecasts generated by the ordinary least squares (OLS) procedure. With the true underlying parameters known in the simulations, estimator basis adn efficiency are both measured. The OLS estimators and predictors are found to be quite robust, generally providing accurate estimates even when OLS conditions are seriously violated. The expanding advocacy of random-coefficient and other stochastic parameter models is open to question. |