We propose predictive jumping (PJ), a fast and efficient algorithm that enables user navigation to off-screen targets. The algorithm is inspired by Delphian Desktop  and the off-screen visualization technique-Halo . The Halos represented at the edge of the viewport help users estimate off-screen target distance and encourage them to make a single fluid mouse movement toward the target. Halfway through the user's motion, the system predicts the user's intended target and quickly moves the cursor towards that predicted off-screen location. In a pilot study we examine the user's ability to select off-screen targets with predictive models based on user's pointing kinematics for off-screen pointing with Halo. We establish a linear relationship between peak velocity and target distance for PJ. We then conducted a controlled experiment to evaluate PJ against other Halo-based techniques, Hop  and Pan with Halo. The results of the study highlight the effectiveness of PJ.