A sentiment-enhanced hybrid model for crude oil price forecasting

Abstract

The crude oil market plays a vital role in the world economy. However, due to the noisy characteristics of the market and the complex and non-stationary nature of the asset series, forecasting the price of oil is particularly challenging. In this study, a new hybrid forecasting approach named FinBERT-VMD-Att-BiGRU is proposed. This integrates FinBERT, variational mode decomposition (VMD), an attention mechanism, and the BiGRU deep-learning model. Specifically, we apply the FinBERT approach to extracting news information for price forecasting, apply VMD to decompose the complex sequence of price series into several simple and stationary subseries, use an attention mechanism to implicitly assign weights to the input features of the deep-learning model, and then adopt BiGRU for price forecasting. The proposed forecasting framework can not only extract qualitative information from crude oil news headlines but also capture both internal and external factors relating to the oil market. Our experimental results show that: (1) the sentiment-enhanced hybrid forecasting approach significantly improves the forecasting performance measured using various benchmarks; (2) the weighting scheme in the sentiment analysis effectively increases the accuracy of the forecasts; (3) a trading strategy based on forecasting results generated by the proposed model can outperform several other common trading strategies. In short, our proposed FinBERT-VMD-Att-BiGRU model has excellent performance in forecasting the price of crude oil.