A recent study by Luciana Juvenal and Ivan Petrella suggests that the financialization of oil futures markets contributed significantly to the surge in oil prices after 2003. Lutz Kilian, Professor of Economics at the University of Michigan, questions their analysis and highlights that their paper actually does not shed any light on the role of Wall Street speculation.
The question of how much speculative pressures contributed to the surge in the real price of oil between 2003 and mid-2008 continues to be hotly debated in policy circles. A common view among policy makers is that excessive speculation driven by the financialization of oil futures markets played a key role in causing oil prices to peak at unprecedented levels in mid-2008. This interpretation has been driving recent policy efforts to tighten the regulation of oil derivatives markets in the U.S. as well as abroad. In sharp contrast, the academic literature on this subject is virtually unanimous that financial speculation played no independent role in this episode. One of the rare studies claiming some success in pinning down the effects of financial speculation has been a working paper by Luciana Juvenal and Ivan Petrella (2012) originally published by the St. Louis Fed in 2011. This study has received considerable media attention, but what does it really show and how were its surprising conclusions arrived at?
Based on estimates of a structural vector autoregressive (VAR) model of the global market for crude oil, Juvenal and Petrella (2012) report that, overall, financial speculation (as captured by what I will refer to as the “hybrid speculative shock” and what they call somewhat imprecisely the “speculative shock”) accounts for about 15% of the surge in the real price of oil from 2003 to mid-2008. This compares with 9% attributed to oil supply disruptions (or “flow supply shocks”) and with 58% attributed to oil demand shocks associated with global business cycle fluctuations (or “flow demand shocks”) – not 40% as has been widely, but incorrectly reported in the media on the basis of the Juvenal-Petrella study (see Figure 1 below).
Figure 1:Juvenal and Petrella’s (2012) Historical Decomposition of the Cumulative Change in the Real Price of Oil Since 2003
These results are presented by the authors as evidence in favor of the view that the financialization of oil futures markets is to blame for speculative pressures on oil prices. To quote Juvenal and Petrella: “The effect that speculation had on oil prices over this period coincides closely with the dramatic rise in commodity index trading – resulting in concerns voiced by policymakers.”  Indeed, their paper prompted headlines such as “Oil Prices Spike Exacerbated by Wall Street Speculation, Federal Reserve Study Finds” (Huffington Post, May 20, 2012), and has been used to justify tighter regulation of oil derivatives markets.
It is fair to say that this study received disproportionate attention in the media because it seemed to produce for the first time solid evidence of financial speculation, reinforcing existing public perceptions. Many casual readers of this study skipped the details and remained unaware of the caveats and assumptions on which Juvenal and Petrella’s conclusions were based. This is not surprising because these details can be complicated, as I know from my own work, which has dealt with the difficult question of how to quantify the role of speculation. In fact, in an earlier structural VAR study with Dan Murphy at the University of Michigan we answered exactly the same question posed by Juvenal and Petrella, but found no evidence that speculative demand shocks (or for that matter financial speculation) caused the real price of oil to increase after 2003 (see Kilian and Murphy (2011)). This raises the question of how the answer obtained by Juvenal and Petrella can be so different. To understand the differences in conclusions, one needs to examine Juvenal and Petrella’s analysis in more detail.
It is useful to start with some background on what we mean by speculation (also see Fattouh, Kilian and Mahadeva (2012)). From an economic point of view, anyone buying crude oil not for current consumption, but for future use is a speculator. Kilian and Murphy (2011) observe that the demand for stocks (or above-ground inventories) of crude oil shifts, as expectations about future demand and supply conditions evolve. We show how to distinguish such speculative demand shocks from shocks to the flow supply of oil or the flow demand for oil in particular. Speculative purchases of oil usually occur because the buyer is anticipating rising oil prices. The same expectations about rising oil prices may also trigger the accumulation of below-ground oil inventories (or reserves), as oil producers choose to keep oil below the ground. In our model such speculative supply shocks are observationally equivalent to flow supply disruptions.
Juvenal and Petrella take the original Kilian-Murphy framework as their starting point. They attempt to extend this model in an effort to distinguish between speculative supply shocks and conventional flow supply shocks, while retaining the speculative demand shock originally introduced by Kilian and Murphy. This is an ambitious and challenging undertaking, and I will show next that their proposed solution has serious limitations.
Juvenal and Petrella’s analysis appeals to a static economic model in which there is an exogenous upward shift in oil producers’ expectations of the real price of oil (possibly driven by the financialization of oil futures markets). Oil producers expecting a higher price for future deliveries, as a result, will withhold oil from the market and accumulate inventories. This accumulation takes the form of an accumulation of below-ground inventories notably in OPEC countries, resulting in a reduction in flow supplies, associated with a higher spot price and lower oil consumption and hence lower real activity. The problem is that, as observed earlier, such a shock is observationally equivalent to any other disruption of flow supplies, say due to civil strife or a strike. This observation prompted Kilian and Murphy (2011) to stress that for all practical purposes speculative supply shocks and flow supply shocks cannot be separately identified (also see the discussion in Hamilton BPEA 2009).
Juvenal and Petrella propose to circumvent this identification problem by imposing the working hypothesis that speculative supply shocks are perfectly correlated with speculative demand shocks such that we can think of these shocks as one “hybrid speculative shock”. This allows them to add the identifying restriction that the impact response of above-ground crude oil inventories in oil importing countries to a negative speculative supply shock is positive. This hypothesis, of course, is not an implication of economic theory. In addition, there are good reasons to believe that for parts of their estimation sample the notion that speculative supply and speculative demand shocks are perfectly correlated does not hold. For example, in 1990 Saudi Arabia as the supplier of last resort increased its oil production following the invasion of Kuwait when it should have cut back production according to Juvenal and Petrella’s model.
Interestingly, adding such a hybrid shock to the model renders the original speculative
demand shock proposed by Kilian and Murphy redundant. Instead of dropping the latter demand shock, as one might have expected, Juvenal and Petrella impose an additional sign restriction on the inventory response to positive flow supply shocks. They insist that this impact response is negative – an identifying restriction for which there is no theoretical rationale, but that is crucial for estimating the model in question.
The problem here is that refiners may respond to a flow supply disruption by drawing down their above-ground inventories (if the supply disruption is considered temporary) or they may choose to accumulate additional inventories (if even higher future oil prices are expected). The latter situation may arise even when the supply disruption does not represent a speculative supply shock. Juvenal and Petrella defend their sign restriction on the grounds that it is consistent with empirical estimates in Kilian and Murphy (2011). That argument is not persuasive given that they also claim that the Kilian-Murphy model is wrong, making it impossible for them to appeal to our empirical results.
This second additional identifying assumption allows Juvenal and Petrella to retain two “speculative” shocks within their model and to distinguish them from the flow supply shock. The modified flow supply shock in their model, of course, now differs substantively from the original flow supply shock Dan and I were studying. To add further confusion, the paper mislabels and misinterprets the original speculative demand shock as an “inventory demand shock” reflecting only shifts in precautionary demand. Both this label and the interpretation of this shock are misleading. Figure 1 reinstates the correct label. Likewise, I have chosen to clean up the remaining shock labels, so they are correct and can be compared with our earlier work.
Even with these (and many other) changes in the model specification, Juvenal and
Petrella’s estimates show that flow demand shocks are the primary cause of the oil price surge after 2003, confirming the substance of the findings in Kilian and Murphy (2011). The key difference is that they find some evidence for the presence of hybrid speculative shocks as well. This raises the question of whether the two models are equally credible and which result to believe. One obvious difference is that the original Kilian-Murphy analysis was based on uncontroversial implications of standard economic models rather than ad hoc assumptions without basis in economic theory. If that is not enough of an argument, we can also judge the credibility of their analysis by inspecting their empirical results.
For example, their paper shows that the hybrid speculative shock not only causes a sustained increase in the real price of oil, but also implies a booming world economy. That finding is robust across all specifications. If this result were taken at face value, it would imply that what is needed in today’s economy are more speculators in oil markets. This is not what the media picked up on as the message of the paper, of course, and the authors wisely do not dwell on this point. Common sense tells us that this result cannot be right, but then this fact implies immediately that the results of the historical decomposition in Figure 1 (which are based on the same impulse response estimates) cannot be correct either, so caution is called for. It is intuitively clear that, without this puzzling feature of increasing global real activity in response to a hybrid speculative shock, the model would assign lower weight to the hybrid speculative shock in fitting the data (and correspondingly even higher weight to the flow demand shock).
One additional reason why these results look strange is that they are not based on a proper econometric evaluation of the model. It is not clear at all how to interpret the impulse response estimates and historical decompositions shown. This general point has been discussed at length in recent work by Inoue and Kilian (2011). It might have been better to show the full range of results consistent with the data even if that dilutes the message of the paper. Yet another reason for skepticism is that Juvenal and Petrella fail to impose several additional identifying assumptions that Dan and I showed to be essential for identification. The failure to impose these restrictions immediately invalidates their model estimates (see Kilian and Murphy (forthcoming JEEA 2012)). How much of a difference this omission makes for the results is hard to tell ex ante, but certainly dropping the impact sign restriction on the response of real activity to the hybrid shock must have important effects on the estimator.
All these caveats suggest considerable caution in interpreting the results. Most importantly, however, their econometric analysis does not tell us anything about whether the hybrid speculative shock is actually associated with financial speculation, as implied by Juvenal and Petrella. Several of the problems I discussed earlier presumably could be fixed or mitigated by more careful attention to the shock identification and model specification, but this problem of interpretation is not one of them. Evidence of “speculation” in models such as Juvenal and Petrella’s merely means that agents expect the real price of oil to increase. It is entirely possible for that speculation to reflect perceptions of higher future scarcity of oil, driven by slower expected growth in world oil production than would be required to meet rising demand driven by emerging Asia, for example. It is also possible that speculative pressures between 2005 and mid-2006 might have arisen because of the positions taken by index fund traders and that those positions were not related to expected economic fundamentals, as suggested by Masters (2008), among others, but there is no way of knowing in the context of this model. Hence, Juvenal and Petrella (2012,
abstract) overreach when they suggest that their “results support the view that … the
financialization [...] of commodity markets [...] played a role … in the recent oil price increase”.
Moreover, there are further arguments against their interpretation. One is that closer inspection of Figure 1 reveals that the hybrid speculative shock increased the real price of oil mainly between 2005 and mid-2006. It played essentially no role during the second phase of rising prices, from mid-2006 through mid-2008, when many pundits suspected particularly large speculative pressures, or between late 2008 and the end of 2010 when index fund positions continued to grow by most accounts. The other argument is that the notion that index funds traders were responsible for rising oil futures prices has been decisively rejected by a large number of recent empirical studies based on alternative methodologies. Those findings are consistent with Kilian and Murphy (2011), but inconsistent with Juvenal and Petrella (2012). The reader is referred to the references in Fattouh, Kilian, and Mahadeva (2012) as well as recent work by Hamilton and Wu (2012).
In short, readers of Juvenal and Petrella (2012) should beware that there is nothing in their analysis that sheds light on the quantitative importance of financial speculation or that would justify tighter regulations for oil derivatives markets. Unfortunately, it is precisely those aspects of their study that have generated the most media attention.
This post written by Lutz Kilian.