Forecasting oil prices: New approaches

This paper proposes alternative methodologies for oil price forecasting using mixed-frequency data and a textual sentiment indicator. The latter variable was extracted from oil market reports issued by the Energy Information Administration. We used the root mean square error (RMSE) to evaluate the forecasting accuracy of the econometric models. Compared with other econometric models, the mixed data sampling (MIDAS) model with high-frequency financial indicators and the sentiment index as explanatory variables performs better for forecasting crude oil prices. •We find the series of oil price has a non-linear quadratic pattern.•We present evidence of the importance of considering data of mixed frequencies.•We point out the benefits of including a sentiment index when forecasting oilprices.•The paper proves that the way of collection and construction of data matter for oil price forecasting.