Learning to anticipate other agents’ future move- ments has gained increased interest in robotics, especially in situations requiring interaction. Such a prediction allows the robotic systems to plan their actions as a task evolves and before its completion. Specifically, in case of active robot collaboration, early decision making might ensure smoothness during motion and collision avoidance. While previous work has addressed prediction and recognition of primitive actions, to our knowledge little attention has been paid to the aspect of unseen ones. In this paper, we introduce a method for online continuous prediction of the evolution of previously unseen trajectories based on a recency weighting of past observations using a library of trained probabilistic movement primitives (ProMPs). The proposed method makes the assumption that parts of any novel trajectory could be considered as a combination of simpler ones. A probability distribution is derived across the evolution of the trajectory, and is updated in a recursive manner. We present a set of simulation experiments to showcase the new method, and compare it with a state-of-the-art method.