Nonlinearity in Global Crude Oil Benchmarks: Disentangling the Effect of Time Aggregation

We model the first and second moments of global crude oil benchmarks, using iterative pre-whitened generalized autoregressive conditional heteroskedasticity (GARCH) models and, in doing so, validate the efficacy of such models in assimilating the neglected nonlinearities in the underlying data-generating processes. The benchmarks considered for this study are Brent, Dubai/Oman, and West Texas Intermediate (WTI) crude oil. While nonlinear serial dependence happens to be a stylized fact across different asset classes, it is our view that prior scholarly contributions have not adequately untangled the effect of data aggregation (in time) in the examination of nonlinear dependencies. In this context, the present study strives to untangle the critical role that time aggregation plays in the examination of nonlinearity in global crude oil benchmarks using data at daily, weekly as well as monthly time frequencies. Our findings are as follows: the optimum GARCH models perform well in capturing all of the neglected nonlinearity in monthly returns of the crude benchmarks. When it comes to daily and weekly returns, our study reveals traces of neglected nonlinearities that are not completely captured by GARCH models. Moreover, such residual traces of neglected nonlinear dependencies are relatively more pronounced at the granular levels and become more and more elusory as the data get aggregated in time. JEL Codes: C22, C53, C58, G1, Q47