Authors: Emmanouil Sofianos, Athanasios-Fotios Athanasiou, Theophilos Papadimitriou, Periklis Gogas

Title: Directional forecast of crude oil daily price using Machine Learning

Abstract

Transportation and industrial products needs oil, so, its price affects the cost of products and inflation. Thus, crude oil price affects the global economy and a method with high forecasting accuracy is of high importance for investors and policy makers.

The aim of the study is to create a directional forecasting model for the WTI (West Texas Intermediate) price of the next day, using the Support Vector Machines (SVM) methodology. The dataset consists of daily prices of WTI and 1, 2, 3 and 4 months Crude Oil Future Contracts, spanning from 09/08/2007 to 01/29/2016 (a total of 2,136 observations).

In the empirical part of our study, we tested both the linear and the RBF (Radial Basis Function) kernel. The model’s parameters were optimized through Cross-Validation Training and Testing framework. The final models were evaluated using the out-of-sample part of our dataset.

The highest prediction performance is achieved using the RBF kernel with a forecasting accuracy of 78.44% while for the logit model the forecasting accuracy reached 73.46%.