Super Spike (GS) -- Finally got a chance to read their report and its really a shame that they gave it the title they did. Once you peel away the headline-grabbing hyperbole there is some damn good work in that piece. Everyone just stopped at the headlines and no one actually read what the guy had to say.
Most notable are Exhibits #5 & 6 and related discussion. Exhibit 5 shows the results of a multi-factor regression model built by GS last December, which predicts the near-month price of crude oil futures by looking at 4 variables: "long-dated" CL futures (i.e. > 60 months out), the spread between light and sour crude, DOE crude & products inventories, and the amount of net speculative length in the Nymex oil futures markets. This model has been extremely accurate at predicting the price of oil, even in the last 4 weeks. Over time it shows an R squared of 85%, which although I'm not a statistician, I believe is pretty impressive.
At Exhibit 6 they show the results of a model that just uses the level of DOE crude and products inventories in an attempt to predict oil prices. This one, while very reliable in the past, has shown to be highly unreliable recently, especially in the last 4 weeks. This model has an R squared of only 52%, which is barely better than a monkey throwing a dart I believe.
As to why the long-dated WTI futures prices are predictive, the report puts forth the theory that these prices are related to market perceptions of the present and future E&P cost structure. At Exhibit 7 they show a graph of E&P upstream costs from '96 to the present, interposed with a graph of the 60 month CL futures price. The 2 seem to move in sync.
So my takeaway from the above is that high finding costs are what is ultimately driving world oil prices, and unless and until these come down (an unlikely event IMO) we will continue to see high oil prices, and this is irrespective of the amount of oil inventories that the US may happen to have at any given point in time.
If anyone else has seen the report and would care to comment I'd welcome your thoughts. Also, if there are any statisticians in the joint could you give me a refresher on the significance of "R squared" when looking at a regression model. Thanks in advance. |
