We investigate whether long-term and short-term components of typical conditioning variables in asset pricing studies, such as the dividend yield or yield spread, have different implications for optimal asset allocation. We argue that short-term components relate mostly to momentum, and long-term components relate mostly to mean-reversion effects, respectively. Therefore, they may have a different information content for investors with different horizons. We obtain improvements in terms of out-of-sample Sharpe ratios and expected utilities for decomposed state variables that directly reflect information related to the stock market, such as the dividend yield and stock market trend.