We show analytically under quite general conditions that implied rates of return based on analysts' earnings forecasts are only a downward biased estimator for future expected one-period returns and therefore not suited for computing market risk premia. The extent of this bias is substantial as verified by a bootstrap approach. We present an alternative estimation equation for future expected one-period returns based on current and past implied rates of return that is superior to simple estimators based on historical returns. The reason for this superiority is a lower variance of estimation results and not the circumvention of the discount rate effect typically stated as a major problem of estimators based on historical return realizations. The superiority of this new approach for portfolio selection purposes is verified numerically for our bootstrap environment and empirically for real capital market data.