The determinants of oil price changes have been extensively investigated in the literature. However, the economic source of its volatility receives much less attention in existing studies. In this paper, we investigate the predictive ability of macroeconomic variables to oil spot price volatility. Using a variety of predictive regressions, we find that some variables such as lagged oil production uncertainty and futures price volatility have significant effects on current oil volatility in-sample. However, adding any predictor to the autoregressive model cannot consistently improve the predictive ability from the out-of-sample perspective. We find the strong and significant predictability using forecast combinations. The revealed predictability is further demonstrated to be robust to the change of lag order and an alternative evaluation criterion of success ratio. Combination methods can provide more accurate density forecasts of realized volatility than the benchmark model. (C) 2019 Elsevier Ltd. All rights reserved.