Abstract (EN)

Overview: The oil-gas relationship is investigated using recent econometric methodology based on a factor analysis and a proper analysis of correlation in residuals. Our methodology allows bypassing the two main issues when dealing with the question of comovement. First, it is important to define what “comovement” is. To do this, we select and estimate factors from a very large number (hundreds) of economic and financial variables. We use these factors to filter out the returns using linear and nonlinear specifications. We then investigate the correlation between the residuals to conclude about a significant comovement between both series of returns beyond what could results from common factors. Again, an issue arises because the return time-varying volatility biases sample correlation (see Forbes and Rigobon (2002) and references therein). We thus adjust the correlation of residuals using the methodology in Forbes and Rigobon (2002), recently applied in Kallberg and Pasquariello (2008) in the framework of sectoral stock indices. Our results do not indicate any decrease in the correlation between oil and gas returns as is suggested in the recent economic literature (see references in the Result section). We can thus conclude that any decoupling between oil and gas prices should be due to common factors leading this relation. Methods : We first use factor models to consider simultaneously the influence of a very large number of economic and financial variables. The methodology is developed in Stock and Watson (2002a and b) and surveyed recently in Bai and Ng (2008). This methodology permits to filter oil and gas returns linearly and nonlinearly so as to remove what could have an impact on both commodities thereby resulting in spurious correlation if any. We construct factors using an international set of economic and financial data, much larger than those used in Stock and Watson and number of papers which used the same data, because we suppose that oil and gas prices should depend on many variables from different countries, particularly very large consumer countries (China). Once returns are filtered considering an optimal number of factors (see Bai and Ng, 2008), we examine the level of correlation between the residuals. It is well-known since at least Forbes and Rigobon (2002) that the coefficient of correlation is biased when volatility is time-varying. Namely, an increase in volatility artificially inflates the level of correlation. As a consequence, we correct the correlation using a moving window volatility adjustment developed by Kallberg and Pasquariello (2008). Results : The correlation between oil and gas returns is not significantly decreasing in contrast with some recent contributions from economic literature (among others, Serletis and Herbert (1999), Ewing et al. (2002), Serletis and Rangel-Ruiz (2004), Asche et al. (2006), Bachmeier and Griffin (2006), Panagiotidis and Rutledge (2007), Hartley et al. (2008), Brown and Yücel (2008)). We do not observe any recent change in the level of correlation. Conclusions : Using the factor models methodology in conjunction with a rigourous analysis of correlation of residuals considering time-varying volatility permits to investigate the question of comovement in oil and gas returns properly. We show that, contrary to what has been suggested in recent literature, oil ans gas returns exhibit a significant residual correlation once influences of many predictors have been considered.