On the relationship between oil and gas markets: a new forecasting framework based on a machine learning approach

Owing to the uncertainty around the coupling and decoupling of oil and gas prices, this study re-examines the relationship between oil and gas markets by modeling the price of one energy source based on the price of the other, both linearly and nonlinearly. We present an autoregressive exogenous model and three nonlinear frameworks with different patterns of asymmetry. Based on daily data from January 7, 1997, to December 29, 2017, our analysis reaches two main findings. First, the nonlinear frameworks outperform the linear model (i.e., the autoregressive exogenous model) in modeling the relationship between oil and gas prices. Second, the nature of asymmetry varies based on market direction. We show that when oil prices exhibit an extreme movement (i.e., beyond a threshold value in absolute value), gas prices react nonlinearly, and that there is no relationship otherwise. Our results are robust for other frequencies, mainly weekly and monthly. These findings explain the conflicting results in the literature on the complex relationship between these markets. The results might serve investors in term of hedging, portfolio diversification, and asset allocation as we show that in the calm period, there is no relationship between oil and gas prices; however, the interaction between markets is more pronounced during periods of extreme movement. Similarly, policymakers’ awareness of the nonlinear dynamic under extreme movements could inform the regulation policy and/or adjustment in case oil (gas) prices increase or decrease.