Modelling Oil Pricing Across Different Regimes: A Neural Network Methodology

Abstract

The global financial crisis of 2007-2009 caused major economic disturbances in the oil market. In this paper we consider five variables describing the microeconomics of supply of and demand for oil and evaluate their importance before, during and after the global financial crisis. We consider five dissimilar regimes during the period of January 1986 to the end of 2017: two regimes prior to the global financial crisis, the regime during the crisis and two regimes after the crisis. The main hypothesis tested is that oil fundamentals of supply and demand remain important even as the five regimes are dissimilar. We build five boosted and over-fitted neural networks to capture the exact relationships between spot oil prices and oil data related to those prices. This analysis shows that, while the inputs into an accurate neural network can remain the same, the impact of each variable can change considerably during different regimes. We find that the shifts in impacts of the various inputs are great enough to support the hypothesis that there are important structural breaks between periods.

Keywords: oil prices, neural network, 2008 crisis, structural break

JEL Classification: D40, E37, Q11, Q31, Q43