The prices of crude oil, gasoline, and heating oil are highly correlated for obvious reasons. In this post I present an estimate of the pairwise linear correlations between the future contracts on these products. The correlations are estimated using a short sample of two weeks. The following figure shows the cumulative return over the sample period where the straight line represents Saturday and Sunday, when the market is closed.
I estimate the pairwise correlation for a range of sampling frequencies from 1 second out to 120 seconds. I use samples of the midpoint to avoid the impact of the bid ask bounce effect. The correlations break down as we approach the highest sampling frequency, which is a well documented phenomenon called the Epps Effect. That said, the correlations remain above 50% at the one second frequency which is evidence for the high frequency spreader algorithms that are trading these contracts at sub second frequencies and in the process strengthening the correlations.
Of course, these correlations evolve over time, and will vary depending on the sample used to estimate. However, the same pattern emerges, that is diminishing correlations with increased sampling frequency.
No comments:
Post a Comment