Estimation of the inputs is the main problem when applying portfolio analysis, and Markov regime-switching models have been shown to improve these estimates. We investigate whether the use of two-regime models remains superior across a range of values of risk aversion and transaction costs, in the presence of skewness and kurtosis and no short sales. Our results for US data suggest that, due to differences in their risk preferences and transactions costs, most retail investors may prefer to use one-regime models, while investment banks may prefer to use two-regime models.