An UNCTAD paper by David Bicchetti and Nicolas Maystre analyzing high frequency data from commodity futures markets has received a tremendous amount of attention in the few days since its release. The paper purports to present evidence that high frequency trading has distorted commodity prices, hence the flurry of attention to what is in fact a rather unexceptional paper. Indeed, the only thing exceptional about it is the authors’ extreme over-interpretation and sensationalization of the results, and their failure to consider seriously more reasonable interpretations of their findings. Their characterization of their findings borders on academic malpractice, and raises questions about their sincerity.
The empirics are rather straightforward. Bicchetti and Maystre calculate rolling correlations between EMini S&P 500 futures returns and the returns on 6 commodity futures contracts-WTI, corn, soybean, wheat, sugar, and live cattle. These correlations are calculated at various frequencies ranging from 1 second to 1 hour.
They find that correlations were typically around zero before jumping substantially around September, 2008. Based on this, they conclude that “the financialization of commodity markets has an impact on the price determination process. . . . In fact, the strong correlations between different commodities and the S&P 500 at very big frequency are really unlikely to reflect economic fundamentals.” They further claim that “HFT strategies, in particular the trend-following ones, are playing a role.”
Their basis for this claim is extremely thin. They note that HFT trading has become increasingly important in recent years. They note that correlations jumped in 2008. They conclude that the former caused the latter.
Gee, now what happened in September, 2008? Let me think. Think. Think. Think. Oh, I remember now-we had a financial crisis!
Could that maybe explain the change in correlations?
Damn right, it could-and did. First, note that the correlations between the S&P 500 and commodity prices are going to be driven primarily by whether supply or demand shocks are driving commodity prices. Take oil as an example. When supply shocks are important, one would expect zero or even negative correlations between oil prices and stock prices. An adverse oil supply shock should lead to higher oil prices and lower stock prices.
When demand shocks predominate, however, positive correlations are to be expected. Adverse aggregate shocks depress both oil prices and stock prices.
Thus, correlations depend on the composition of shocks. When demand shocks (driven by shocks to income) predominate, correlations should be positive; when commodity-specific supply shocks predominate, correlations should be zero or negative. (With the ags, since supply shocks are unlikely to be correlated with aggregate income or growth, one would expect near-zero correlations. With oil, supply shocks can have macro effects, so negative correlations can arise.)
Starting in the fall of 2008 there was a dramatic increase in the flow of information about demand-related shocks. The predominant feature of those months was a dizzying flow of macroeconomic-related news. Would major economies be thrown into depression? Would the financial sector collapse? Would economies recover? How rapid would the recovery be? What were governments doing to trying to stem the collapse and rekindle growth? The composition of economic shocks shifted decisively towards demand/income-related shocks that dwarfed commodity-specific supply-related shocks.
In this environment, commodity-specific supply-related information was swamped by systematic demand-related information. You would expect-clearly-this to lead to positive co-movements between stock prices and commodity prices, because both were being driven by demand related information.
Moreover, financial shocks during the crises that constrained risk-bearing capacity that were quite prevalent during this period would have affected both stocks and commodities, and in the same way. These shocks to risk bearing capacity would have affected expected returns on commodities and equities in the same way. This would have also contributed to substantial positive covariation.
The very discontinuity in the correlations around the time of Lehman supports this view. Algorithmic trading and HFT trading (they are different) were both growing throughout the mid-to-late-2000s. There was no discontinuous jump in the utilization of these strategies at the time of Lehman. But there was a discontinuous jump in the correlations. A much more plausible interpretation of the data is that a discontinuous change in the nature of economic shocks (becoming predominately economy-wide demand-related shocks) and correlated shocks to risk bearing capacity (that led to correlated shocks in expected returns) attributable to the (discontinuous) financial crisis explains the discontinuous change in correlations. The authors offer no plausible explanation-none-of how a continuous growth in HFT could lead to a discontinuous jump in correlations precisely in September, 2008.
Indeed, their measure of HFT-the ratio of volumes to trades-flattened out precisely during this period, after having fallen since the beginning of 2007. If this is a measure of HFT, it should have jumped down discontinuously at the time correlations jumped up if in fact HFT was driving correlatons. But the exact opposite happened. The authors let this obvious problem with their interpretation pass in silence. This is inexcusable.
There is other information in the report that strongly supports the view that fundamentals were in fact driving the correlations. In particular, the WTI-S&P correlation dropped substantially in January-April, 2011. This is precisely the period in which the Arab Spring and in particular the outbreak of civil war in Libya led to substantial supply shocks in the oil market. As noted above, supply shocks for oil should lead to opposite movements in oil prices and stock prices. An increase in the flow of supply-related information should lead to lower correlations between stock and oil prices. The Arab Spring and Libyan events led precisely to such an upsurge in supply shocks, and correlations fell exactly as would be expected if fundamentals were actually driving prices. Under the HFT interpretation, there would have had to have been a drop in HFT utilization during this period, which certainly did not occur.
In sum, the evidence presented in the widely hyped Bicchetti-Maystre paper in fact strongly supports the view that fundamentals were driving commodity and stock prices during the sample period. In particular, the main piece of evidence that B-M emphasize-the jump of correlations in September, 2008-corresponds to an obvious regime change in which economy-wide shocks that would affect both stock prices and the demand for commodities occurred more frequently and with greater severity. One would predict-and in fact I did so predict contemporaneously-that such a change would lead to a substantial change in the correlations between stock and commodity prices. (I made this prediction publicly at a symposium held at the Bauer College shortly after the beginning of the crisis, and actually made the conditional prediction in the summer of 2008 that the only thing that would bring down oil prices was a serious economic downturn.) The change in oil-stock price correlations around the time of the Libyan crisis is also consistent with the hypothesis that price movements reflected information about fundamental supply and demand conditions.
In contrast, there were no discontinuous changes in “financialization” or the utilization of HFT either in September, 2008 or during the Arab Spring/Libyan period.
In sum, the actual evidence presented in the paper strongly supports the hypothesis that fundamentals drive prices. It does not provide any support for the hypothesis that “financialization” or HFT is distorting prices and overriding the impact of fundamentals. Indeed, there is strongly contrary evidence.
Thus, although the empirical analysis in this paper is unobjectionable, the interpretation is contrary to the actual results and a reasonable reading of the economic and financial history of this period. Indeed, the interpretation is so at odds with the data and history that the motives of the authors are open to serious question. They jump on the anti-financialization, anti-HFT bandwagon even though (a) the actual data does not support that interpretation, and (b) other, quite opposed, interpretations are much more reasonable. Whenever there is a yawning gap between the most reasonable interpretation of an empirical analysis, and the interpretation that the authors advance, one may reasonably question the intellectual honesty of those interpretations.
And I do so question.
I therefore recommend examining the various graphs presented in the paper in light of an understanding of how the composition of economic shocks should affect correlations, and an understanding of what was happening during the sample period of the study. I further recommend completely disregarding the Bicchetti-Maystre interpretation of their results, because it betrays no understanding of either the economics or the history.