Is the refining margin stationary?

It has traditionally been assumed that the refining margin is stationary given that it is a linear combination of cointegrated time series, i.e., crude oil and its main refining products (mainly heating oil and gasoline). Following this reasoning, stationary models have been proposed to measure the refining margin. In this paper, we investigate the main empirical properties of several time series that measure the refining margin (or crack spread) using an extensive database of WTI, heating oil and unleaded gasoline futures prices traded on the NYMEX. The results show that there are serious doubts about the stationarity of the refining margin. Moreover, a non-stationary factor model is proposed and estimated to measure the refining margin, and in some cases, the model achieves better results than the traditional stationary models. This result has straightforward implications for valuation and hedging.