We analyse the generalized, dynamic Nelson-Siegel approach including five factors to ensures the absence of arbitrage. In contrast to previous empirical analyses we define our risk factors so that they are observable and determine their significance using a series of cross sectional regressions. We then decompose the risk factors and test whether they are significant on each time scale. The results allow us to distinguish expected and unexpected components which are used in out of sample forecasts. We find good forecasting abilities of this approach; the one month forecast remains high even during times of financial market distress.