In recent papers, Cecchetti et al. (1990) and Kandel and Stambaugh (1990) showed that negative serial correlation in long horizon returns was consistent with an equilibrium model of asset pricing. In this paper, we show that their results rely on misspecified Markov switching models for the endowment process. Once the proper Markov specification is chosen for the endowment process, the model does not produce mean reversion of the magnitude detected in the data. Furthermore, the small amount of mean reversion produced by the model is due only to small sample bias. We also show that this model is unable to predict negative excess returns, contrary to empirical evidence.