This paper illustrates the importance of density forecasting in portfolio decision making involving bonds of different maturities. The forecast performance of an atheoretic and a theory informed model of bond returns is evaluated. The decision making environment is fully described for an investor seeking to optimally allocate his portfolio between long and short Treasury Bills, over investment horizons of up to two years. Using weekly data over 1997 to 2007 we examine the impact of parameter uncertainty and predictability in returns on the investor's allocation. We describe how the forecasts are computed and used in this context. Both statistical and decision-based criteria are used to assess the out-of-sample forecasting performance of the models. Our results show sensitivity to the evaluation criterion used. In the context of investment decision making under an economic value criterion, we find some potential gain for the investor from assuming predictability.