Forecasting Daily Nigerian Crude Oil Prices using a Hybrid 1D CNN LSTM Network

Abstract:

The economic growth and well being of citizens in all countries of the world depend on the price and steady supply of crude oil. Crude oil and its refined derivatives play an important role in driving the global economy and providing citizens of all countries of the world an easy and comfortable lifestyle. Hence, when supply of crude oil is threatened, or its price rises abruptly, government officials, traders, and manufacturers become deeply concerned and try to regularize its supply and demand and stabilize its price. The dynamics of supply and demand, economic growth and lifestyles have made the price of crude oil highly irregular, uncertain, and difficult to forecast. Researchers and academics worldwide have developed and continue to develop new models to forecast accurately the price of crude oil. An accurate model will allow energy policy makers, investors, traders, and manufacturers to prepare and mitigate the impact of rising prices both on countries that depend solely on imported crude oil and countries that are sustained by its export. This work develops a 1D
CNN LSTM hybrid model for forecasting the daily price of Nigerian crude oil and compares it performance with an LSTM model. The 1D CNN LSTM hybrid model outperforms the LSTM model thereby validating our decision to develop the hybrid model. 1D CNN improves the forecasting capability of the hybrid model by extracting spatial and short term features of crude oil price fluctuations and passing them to LSTM to learn the long term dependencies and correlations of these features.

Keywords: Crude Oil Price Prediction, Deep Learning, Long Short Term Memory (LSTM), Convolution Neural Networks (CNN)