Short answer: not much. But here are some fun facts to know and tell that do tend to deflate any grandiose assertions about the effects of speculation on oil prices:
- Changes in non-commercial positions did not Granger Cause crude oil returns in the 1986-2009, 2001-2009, or 2006-2009 periods in bivariate Granger Causality analyses. Not that this is a surprise; if one could predict future oil price changes based on the publicly released COT numbers, one could make money. But it does belie the Medlock-Jaffe claim that non-commercial positions are a “leading indicator” for oil prices.
- In a vector autoregression including crude oil returns, the percentage change in the Baltic Freight Index, the percentage change in the dollar index, and the change in non-commercial positions, the non-commercial positions do not Granger cause the crude oil returns in the 1986-2009, 2001-2009, or 2006-2009 periods. The percentage change in the BFI does Granger Cause crude oil returns in all three samples. This might be surprising, at it would seem to suggest that one can make money in oil by making trades based on moves in the BFI. However, it is more likely that the movements in the BFI forecast time varying expected returns. The BFI is a proxy for industrial activity, and thus likely reflects systematic risk.
- In the entire sample, crude oil returns Granger Cause (i.e., forecast) changes in non-commercial positions. This is not true in the 2006-2009 period.
- In the entire sample, commercial positions Granger Cause non-commercial positions, but the reverse is not true. In the 2001-2009 and 2006-2009 periods, the causality runs both ways.
- In regressions of crude oil returns against percentage changes in the BFI, percentage changes in the dollar index, and changes in non-commercial positions, all three explanatory variables are statistically significant in the entire sample and the two later subsamples. Changes in commercial positions are associated with changes in crude oil prices of the same sign.
- The magnitude of the coefficient on the change in non-commercial positions is relatively small, and has been declining over time. In the entire sample, a one standard deviation move in the change in the non-commercial position is associated with a 2 percent change in the crude oil price.
- However, by 2006-2009, the alleged period of speculative excess, the coefficient on the non-commercial position change is less than half its value in the entire sample. During 2006-2009, a one standard deviation move in non-commercial positions (corresponding to about 17 mm bbls of crude in a week), is associated with a less than one percent move in the crude oil price. Given that the standard deviation of returns (this is weekly data) during this period was over 5 percent, this is a relatively modest association.
- Moreover, speculative positions rose and fell throughout the period of alleged speculative excess. Using the estimated coefficient on the change in the non-commercial position from the multivariate regression, and the movements in non-commercial positions from the beginning of 2006 to the height of the oil prices in early-July, 2008, the cumulative effect of speculative trading on the price of oil was . . . wait for it . . . 2.56 percent. Not 25.6 percent. 2.56 percent. And this over a period when the price of oil rose from $63 to over $141 dollars. That means, that speculation might have “caused” the price to move from $63 to not even $65. Big whoop.
- The correlation between changes in non-commercial positions was .36 in 1986-2000, .5 from 2001-2005, but fell to .3 in 2006-2009.
- Thus, the data do not support the contention that speculation–at least as measured (or mis-measured) by the COT have played an increasingly important role in affecting oil prices. Indeed, the association between non-commercial futures trading and oil prices was weakest during the period of time of the purported speculative Bacchanal.
- Conversely, the relationship between the crude oil price and the dollar index has strengthened over time. In the 1986-2000 period, the coefficient on the dollar index in the multiple regression was -.023; during the 2001-2006 period, -.77; and from the 2006-2009 period, -2.03. During this last period, a one standard deviation move in the dollar index is associated with approximately a 2.5 percent move in the price of crude oil.
This all took about 40 minutes to figure out.
So what does it all mean? Well, even without grappling with the causation issue, the data provide virtually no support for the hyperventilating assertions that speculation dominates oil prices, and that this domination grew over time, reaching a frenzied peak in 2006-2008.
In contrast the data provide better support for the view that the price of oil became more dollar driven. Thus monetary policy should deserve more scrutiny, and speculation less, in trying to understand what happened to oil prices. The attempt by Medlock-Jaffee to attribute the decline in the dollar to oil speculation is implausible.
But even though there is a modest, contemporaneous association between non-commercial positions and oil prices, that provides only the weakest support for the view that speculation distorted prices. There is the correlation-causation issue. What caused the changes in non-commercial positions? Could they have moved in response to oil prices? Could both have been driven by common factors? Moreover, even if speculation moves prices, it can move prices towards where they should go. That is, informed speculation tends to move prices towards their full-information value.
As I’ve written often, looking at prices alone is often inadequate to diagnose speculative distortions. Prices are important because they provide signals that affect the allocation of resources. That is, prices affect decisions about quantities. If prices are distorted, quantities should be distorted too. For the commodities allegedly affected by speculative excess in the 2006-2008 period, there is no evidence that this was the case. In fact, for the commodities for which the data are best (industrial metals, where daily inventory data are available), the evidence is quite to the contrary. The inventory data provide strong evidence of fundamental tightness, rather than speculative excess, as the cause of high prices. (This will the subject of a chapter in my forthcoming book.)
The COT data are also dubious for a variety of reasons, including their crude categorization of traders, their lumping together of positions in all delivery months, and the fact that they cover only exchange-traded positions, and not the OTC market.
So, bottom line. The association between speculative trading (as measured by the COT) and oil price movements is weak. Given this association, and the movements in non-commercial trading during the 2006-2008 period, the cumulative impact of speculation on the price of oil during the spike was miniscule. Moreover, one cannot conclude that the measured association quantifies a causal relationship, and even if it was causal, you can’t conclude that it caused prices to move away from a level justified by fundamentals. The non-price (i.e., quantity) evidence for price distortion is lacking.
In one sentence: The COT data that have been used repeatedly to support contentions that speculation distorted oil prices provide no evidence whatsoever that speculation in the 2006-2008 time period materially contributed to the dramatic rise in the price of oil during this period.