In this paper, we theoretically and empirically study the intra-horizon value at risk (iVaR) in a general jump-diffusion setting. We propose a new class of models of asset returns, the displaced mixed-exponential model (D-MEM), which can arbitrarily closely approximate finite-and infiniteactivity Lévy processes. We then derive analytical results for the iVaR and disentangle, in a theoretically consistent way, the jump and diffusion contributions to the intra-horizon risk. We estimate historical and option-implied VaR and iVaR for several popular jump models using the S&P 100 index and American options. Empirically disentangling the contribution of the jumps from the contribution of the diffusion, we conclude that jumps account for about 90 percent of the iVaR on average. Our backtesting results indicate that the option-implied estimates are much more responsive to market changes than their historical counterparts, which perform poorly.Keywords: value at risk, intra-horizon risk, displaced mixed-exponential model, first-passage disentanglement, option-implied estimates.JEL classification: G01, G11, G13, C51, C52.3 Finally, we expand on the empirical analysis of Bakshi and Panayotov (2010) by providing a backtesting study for historical and option-implied VaR and iVaR figures across different models.While we do find differences of iVaR and VaR violations across different models, these differences are small compared to those between historical and option-implied VaR and iVaR figures. Our backtesting procedure shows that, irrespectively of the model used, the option-implied VaR and iVaR estimates are considerably more perceptive and responsive to asset price fluctuations, and they produce more accurate results than do the historical estimates. For this reason, we conclude that whenever options data is available, the option-implied estimates of risk measures provide additional information that should not be neglected.The paper is structured as follows. In Section 2, we introduce the D-MEM class of exponential Lévy models and connect the VaR and iVaR to the payoffs of European and one-touch digital puts. In Section 3, we describe the data treatment and summarize the calibration and the models' performance results. Our empirical findings for the VaR and iVaR are discussed in Section 4. We conclude in Section 5.