We propose a parsimonious factor model for jump intensities for a large number of stocks, which we call the Jump Factor Intensity Thinning (J-FIT) Model. The model utilizes a jump factor process to capture the complete grid of potential jumps, and a thinning procedure to capture individual stock jumps. The jump factor process features temporal dependence, which leads to predictability. We develop estimators and inference theories for the model parameters, enabling applications such as jump intensity prediction and constructing prediction intervals for the jump risk exposures of individual stocks and sectors. Empirically, we provide solid evidence that jumps are predictable with the J-FIT model, whose predictive performance substantially sur- passes various benchmark models. We also demonstrate that J-FIT is useful for tail risk management.