Recently, Ben Bernanke wrote a blog post that attempted to quantify how much of the oil decline can be attributed to global demand factors. The model builds upon the work of James Hamilton who estimated a regression where changes in oil are determined by changes in copper prices, the 10-year treasury yield, and the dollar. Comparing the predicted and actual decline in oil prices, Ben Bernanke finds that something in the range of 40-45 percent of the actual decline in oil prices since June 2014 can be attributed to unexpectedly weak demand. DavidBeckworth, former economist at the U.S. Department of Treasury, also wrote a blog post where he reestimated the model, but added the BAA minus the 10-year treasury spread as another indicator of global demand conditions. Using his specification of the model, he was able to attribute 51 percent of actual decline in oil prices to unexpectedly weak global demand. Bernanke used daily data since mid-2011, while Beckworth since January 2007.
Here I’m going to tackle the same question. How much of the oil decline can be attributed to global demand factors? I’m tired of hearing that the most important factor has been surging U.S. oil production. But I’m going to do so in the most elegant way I can think of within the same framework. So I’m going to use only copper prices and the 365 day lagged spread between the copper and oil price. I used daily data since 31 January of 2000. Using my model which includes an error correction term, I’m able to attribute 64 percent of the actual oil decline (49.4%) since June 2014 to December 2014 to global demand shocks. This is, even if we knew nothing on the production side of the oil market, we would have anticipated the price of oil to have fallen 32% in December 2014 since June 2014 given unexpectedly weak global demand and past deviations of the spread between the oil and copper price from equilibrium. In relation to the actual oil accumulated decline (64.8%) in December 2015 since June 2014, 54 percent is due to systematic shocks (as opposed to oil supply shock) according to the model.
I resuscitate Engle and Granger (1987) to tackle this question. My approach is based on the fact that the oil and copper price are cointegrated, so past deviations from equilibrium of the spread between them influence today’s oil and copper behavior. So there is useful information in the past spread that we should use to predict present oil change. It makes no sense knowing that the oil and copper share an equilibrium relationship so they tend to make up for past deviations from equilibrium and not incorporating past deviations from equilibrium to form expectations on oil changes.
The procedure is the following: First I estimated the cointegrating equation using data since 31 January of 2000 to June 2014.
Finally, I used the coefficients from that equation to see how much of the drop in oil prices since June 2014 we would have expected to observe, given the observed drop in copper prices and past deviation of the spread from equilibrium. Only the fraction of the observed declined in oil price that can’t be explained by this model, may be attributed to supply oil shocks. On average, I was able to attribute 60 percent of the actual accumulated decline in oil prices during 2015 since June 2014 to unexpectedly weak global demand . As you can see, there is no evidence that the most important factor explaining the oil fall has been surging oil production. Supposed, right now I’m standing in June 2014 and Marty Mcfly (just someone from the future) reveals to me ONLY the copper price time series, an indicator of global nominal spending conditions, since July 2014 to December 2015; the dotted red line in the following graph represent the expected path for the oil price I’m able to come up. Now supposed an alternative future where the oil price in Dec.2014 declined only by 20% since June 2014. Now you see that it makes no sense to argue that the oil decline was due to a positve oil supply shock, since I was expecting it to fall further (by 32%).
|Author’s calculations based on ec.2|
Data downloaded from Bloomberg.