The paper is concerned with the question of whether the pricing of US stocks has been efficient in terms of the present value model. The MTAR model used in the context of market efficiency is extended by means of a rolling window estimation strategy. This rolling MTAR analysis revealed that the assumed underlying time series process in the logarithm of earnings to stock-price ratio is highly unstable over time. The findings cast doubt on the usefulness of conclusions relating to extended time periods of about 130 years. Because of the rolling MTAR approach, it is possible to reveal decades of inefficient stock pricing as well as decades of assymetrical time series behaviour.