Using a temporal convolutional network to predict commodity futures under uncertainty
Abstract
Using a temporal convolutional network to predict commodity futures under uncertainty
Incoming article date: 08.06.2023The article discusses commodity futures price forecasting using a temporal convolutional network. Commodity futures forecasting is an important task for investors and traders because it allows you to predict future prices and the direction of the market. Commodity futures forecasting can be done using a variety of methods and approaches. One such approach is the use of deep learning models, which consists in predicting futures quotes using artificial neural networks. There are many types of neural networks, among them the most popular for the task of processing time series are recurrent neural networks. However, recurrent neural networks have certain disadvantages that a temporal convolutional network does not have. The temporal convolutional network architecture has unique features such as parallel processing of data, extraction of short- and long-term dependencies, and extraction of important features on different time scales. An experiment was conducted to assess the accuracy of predicting the closing price of seven commodity futures using a temporary convolutional network and an ARIMA statistical model with automatic selection of parameters. As a result of the experiment, it was revealed that the temporary convolutional network is superior to the statistical ARIMA model and is a very effective model for forecasting commodity futures. However, despite the high potential of the proposed forecasting model, it is also important to take into account various other analytical methods, such as fundamental analysis and expert opinion.
Keywords: machine learning, temporal convolutional neural network, commodity futures forecasting, commodities, financial time series