No one working on a hot-button issue such as speculation in oil markets could have been so blue-eyed to think that a study reporting “a significant effect of financial speculation” related “to the concerns voiced by policymakers” would not be picked up by the media. In response to my critical review of their analysis and conclusions, Juvenal and Petrella in their rejoinder suggest that their findings have been misunderstood and are largely in line with a large strand of the literature, including my own contributions, but that is not even half true.
Recall that Juvenal and Petrella’s conclusion is (a) that flow demand for oil caused "less than half" of the run-up in the real price of oil, and (b) that "speculation played a significant role" as well. The conclusion in Kilian and Murphy (2011), in contrast, is that flow demand for oil caused nearly all of the run-up in the real price of oil and that speculation played no role at all. Moreover, Juvenal and Petrella’s conclusions regarding financial speculation are a minority view among experts in the field (see Fattouh et al. 2012).
Has the press treated Juvenal and Petrella unfairly?
Juvenal and Petrella further suggest that I unfairly highlighted one "inaccurate" media report in the Huffington Post on their work. There actually are many such reports and they do not differ materially. Another typical example is The Economic Populist’s review of this paper under the title "The Wall Street Surcharge on Your Gas Tank".
In their rejoinder, Juvenal and Petrella seem to suggest that the media are to blame for misrepresenting their paper. Indeed, no one – including myself – would want to hold them responsible for misleading summaries of their work in the press. However, this is not a matter of the media not reporting the results accurately. Rather the problem is that the authors themselves couched their message in words that were misleading. For example, when I summarized the authors as saying that "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" I was citing the authors’ own Economic article, not some blog.
The same goes for the estimates cited in blogs. Juvenal and Petrella cite one particular blog article as having gotten the message right, but did it? The article in question attributes 15% of the cumulative increase in the real price of oil since about 2003 or 2004 to their hybrid shock and 44% to the flow demand shock. While I can replicate the 15% estimate, the actual contribution of the flow demand shock to the total cumulative increase is higher than reported. In my original guest blog I showed the fraction for the flow demand shock to be 58% under their specification, not 44% as reported. That latter figure refers to the cumulative increase (which by my account is 43% to be exact), not the share in the total cumulative increase discussed in the blog. In other words, the blog overstates the importance of speculation relative to flow demand.
Where does the lower number for flow demand come from? It actually is consistent with the 45% figure explicitly mentioned by Luciana Juvenal in a Bloomberg radio interview, in which she also states that “almost half” of the total cumulative increase in the real price of oil can be attributed to flow demand. It is not surprising that bloggers simply round that figure to 40%. A case in point is the The Economic Populist.
In fact, I find the blogs that I have seen (including the Huffington Post) to be remarkably faithful to the working paper. The key statements are taken almost verbatim from Juvenal and Petrella’s work. It seems disingenuous for the authors to blame the press in this matter. My discussion was based on direct quotes from the authors, not on the content of blogs. The only thing misleading here is the interpretation of the results in the original working paper.
Now let me turn to the specific replies on more technical points and explain why I am not persuaded by their analysis including the additional points raised in the rejoinder:
Juvenal and Petrella postulate that speculative demand shocks and speculative supply shocks are perfectly correlated such that we can define one hybrid speculative shock. When I said that this premise does not follow from economic theory, I did not mean that one cannot write down a model that imposes this premise, but that conventional economic models do not imply this restriction. Thus, at best this assumption is a working hypothesis and one has to ask oneself whether this premise is empirically reasonable. This can be done ex ante (I registered some doubts in this regard) or ex post on the basis of the results of the model (I pointed out some strange implications of the estimates). I also noted that implementing this approach requires additional identifying assumptions without any support that go beyond postulating the existence of a hybrid speculative shock.
So what about the theoretical underpinnings that the authors refer to? Contrary to Juvenal and Petrella’s claim in the rejoinder, the particular model in Appendix E actually is not a theoretical representation of their structural VAR model at all. In fact, it lacks all the essential features of the structural model that need explaining. For example, there are no demand and supply shocks of any kind and variables that are endogenous in the VAR model are treated as exogenous in the theoretical model. It is not clear to me what this model adds, except perhaps confusion if we are not careful.
I agree on the substantive point that - if a hybrid shock exists - this shock embodies a simultaneous negative speculative supply shock and a positive speculative demand shock, but that is nothing more than a restatement of the definition of this shock. Neither demand nor supply is observable. The real question is what the observable implications of this shock are. Although the theoretical model presented by the authors does not shed light on this question, the three sign restrictions the authors impose seem uncontroversial, conditional on their premise. The lack of a restriction on the response of real activity to the hybrid speculative shock, in contrast, is controversial, and it is precisely in this regard that the economic model in Appendix E has nothing to contribute. Nor does their economic model provide the other restrictions required for the authors’ results.
2. The 1990 Oil Price Shock
Juvenal and Petrella obtain the result that the 1990 oil price spike, following the invasion of Kuwait, was driven by what they call an inventory demand shock. The first point of confusion is that this is not an oil inventory demand shock, but a speculative demand shock as defined in the Kilian-Murphy model. A speculative demand shock includes as a special case, but is not restricted to, the precautionary demand shock described in Alquist and Kilian (2010). The 1990 episode illustrates the difference. This shift in speculative demand actually was not so much driven by increased uncertainty, as claimed by the authors, as by the expectation of an invasion of Saudi Arabia which triggered an anticipation of lower Saudi oil supplies. In other words it is primarily a first moment shock rather than a second moment shock, as discussed in Kilian and Murphy (2011).
It is not surprising that this event is not classified as a hybrid speculative shock in the Juvenal-Petrella model, precisely because Saudi oil production increased following the invasion, as noted in my original guest blog. What Juvenal and Petrella fail to explain is why oil producers would have failed to respond to this opportunity for speculating on future price increases by curtailing oil production now. After all the economic mechanism they have in mind applies to this episode as well. In particular, the term structure of oil futures in 1990 indicated arbitrage opportunities for oil producers. If they did not respond in this fashion in 1990, what makes us think that they would have after 2003?
3. Is the oil inventory demand shock redundant?
Juvenal and Petrella write that a shift in inventory demand (meaning in speculative demand) need not be matched by a contemporaneous shift in supply. No one suggested that it would. Of course, an exogenous shift of the demand curve along the supply curve does not shift the supply curve – not in Alquist and Kilian, not in the econometric model of Kilian and Murphy and not in any other theoretical model I know of. The question is instead what the effect of a speculative demand shock is on observable oil production in the impact period. Kilian and Murphy note that a positive speculative demand shock in equilibrium raises the real price of oil and the level of oil inventories. This is accomplished by reducing oil consumption and raising oil production on impact. While observable production shifts, the supply curve stays put.
Now to the more central point of whether the speculative demand shock is redundant, given the hybrid speculative shock: There are two modeling strategies here. One is to try to model speculative supply shocks and speculative demand shocks separately. This is not possible, as discussed in my original blog. The second strategy is to make the case that both shocks occur at the same time by construction, resulting in what I call the hybrid speculative shock. This is what Juvenal and Petrella propose. Given that the short-run supply increase in response to a positive speculative demand shock is likely to be weak, the speculative supply shock will plausibly cause a net decline in oil production on impact. I have no problems with that restriction, conditional on the premise of the authors (although I am not sure I buy the premise of the hybrid shock).
Once this hybrid shock is defined, we have already incorporated speculative demand shocks. If there were “other” speculative demand shocks, why would they not be correlated with the speculative supply shocks we already modeled? Is the economic mechanism any different? Juvenal and Petrella seem to think that uncertainty-driven precautionary demand shocks are fundamentally different from other speculative demand shocks, but why? The earlier example of the 1990 episode illustrates why that distinction does not make sense (see point 2). In fact, the identifying restrictions Juvenal and Petrella use for this shock are identical to those for a speculative demand shock in Kilian and Murphy, further reinforcing the point that this shock is not just about precautionary demand shifts in response to exogenous changes in uncertainty. Moreover, if there are speculative demand shocks different from the hybrid shock, why are there no speculative supply shocks that are distinct from the hybrid shock? It seems strange to have one and not the other.
My point is not only that the authors have not thought through the interpretation of their model or that the structure of their model is ad hoc, but more importantly that the model they work with is not properly identified. Recall that a speculative supply shock that is not correlated with speculative demand shocks would have the same sign restrictions as the original flow supply shock in the Kilian-Murphy model. Nor could such a shock be distinguished from the new hybrid speculative shock. In other words, without further restrictions the model of Juvenal and Petrella would not be identified. That’s why the authors add an additional negative sign restriction on the impact response of oil inventories to a negative flow supply shock. Their justification for this ad hoc restriction is that Kilian and Murphy (2011) obtained a negative estimate for that parameter without imposing any restriction. As I pointed out in my original blog that argument is not compelling because their model is different from ours in several dimensions.
As to whether Juvenal and Petrella implied that the Kilian-Murphy model is wrong, the issue is quite simple. My point here is that their identifying restrictions are logically inconsistent with ours, so if one set of identifying assumptions is right, the other one cannot be right at the same time (whether Juvenal and Petrella say so explicitly or not). Hence Juvenal and Petrella cannot borrow empirical evidence from the Kilian-Murphy model and impose it as identifying restrictions in their econometric model. This is important because without this additional restriction, which they cannot motivate based on economic theory, identification in their model is lost. It is true that their work builds in many ways directly on ours, but that is not enough to justify this identifying restriction, rendering their analysis invalid.
4. Differences in approach
Juvenal and Petrella suggest that their theoretical model encompasses the model of Alquist and Kilian (2010) as a special case. Their model postulates an exogenous shift in the future oil price. Unlike in Alquist and Kilian there is no oil futures price in the model. Alquist and Kilian’s model allows for all their model variables including the oil futures price to respond endogenously to structural shocks, so it is hard to see what the basis for Juvenal and Petrella’s statement is. In fact, Juvenal and Petrella’s model does not even allow one to study the effect of an increase in uncertainty about future oil supply shortfalls. Nor does their model allow an analysis of other structural demand shocks of any kind.
In any case, the Kilian-Murphy econometric model encompasses the theoretical framework of Alquist and Kilian, without being restricted to that framework. As to the theory underlying that model, Juvenal and Petrella do not seem to have an issue with copying most of our identifying assumptions, so presumably they are not concerned with the validity of the underlying economic theory. What is controversial about their approach is the imposition of additional restrictions that do not come from either our economic model or their economic model.
5. Speculation and real economic activity
The authors take issue with my statement that a hybrid speculative shock is associated with an increase in global real activity, which in turn implies that more speculation would stimulate the global economy. That statement is merely restating what Figure 2 in their paper shows. Juvenal and Petrella respond that they did not spell out this implication in their paper. Indeed they did not, but they should have done so because that implication is there for everyone to see.
Particularly troubling is the positive impact response of real activity. It is true that the authors did not impose this result, as they insist, but this result nevertheless emerges from their model. In fact, from a mechanical point of view, this result occurs because the authors dropped the impact sign restriction on the response of real activity to the hybrid speculative shock. It is well known that not imposing a required restriction such as the negative sign restriction on the impact response of real activity to a hybrid shock will render the impulse response estimates misleading. In fact, there is a forthcoming paper by Kilian and Murphy (JEEA 2012), which discusses why remaining “agnostic”, as the authors like to call their approach in the rejoinder, is a bad idea when working with VAR sign restrictions.
Juvenal and Petrella in their defense suggest that they need to dispense with this sign restriction because financial speculation may be caused by expected real interest rate shocks, but that argument is not persuasive. In support of this conjecture, they cite Frankel’s work on the role of real interest rates for commodity prices, leaving out conveniently that Frankel and Rose’s more recent empirical work was unable to document such feedback. Nor did Alquist, Kilian and Vigfusson (2012, forthcoming: Handbook of Economic Forecasting) find evidence of a link between interest rates and oil prices. Moreover, Dan and I explicitly reported testing for whether ex ante real interest rates Granger cause the remaining VAR variables. There was no evidence of Granger causality. This negative finding is also in line with more structural evidence in Kilian and Vega (REStat 2011) that exogenous news about interest rates do not move the price of oil, contrary to Juvenal and Petrella’s conjecture. The latter result directly refutes their point.
It is also worth adding that Juvenal and Petrella in their discussion glance over the fact that their theoretical analysis in Appendix E does not allow for shocks to the expected real interest rate. Most importantly, their proposed explanation overlooks the fact that the expected real interest rate is an endogenous variable. Shocks to this observable are not structural shocks. In short, the authors’ discussion mixes up structural models designed to detect causal links and reduced-form models designed to capture correlations.
6. Speculation in the last decade
Juvenal and Petrella wrote in their working paper that: “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.” In their rejoinder, they concede my point that there is no obvious correlation between the volume of index trading in oil futures markets and the explanatory power of the hybrid speculative shock. On this point we seem to be in agreement.
They instead stress the role of the oil futures market being in backwardation or in contango for the explanatory power of the hybrid speculative shock. It is useful to look at this argument more closely. Juvenal and Petrella refer to Singleton (2012, Figure 3) who reports that based on three-month futures contracts the oil futures market was in backwardation from April 2002 until November 2004 and from July 2007 until May 2008. It was in contango from November 2004 until July 2007 and from May 2008 until October 2009. Juvenal and Petrella imply that backwardation means the absence of hybrid speculation, whereas periods of contango imply an increase in the component of the real price explained by the hybrid shock.
So what do their model estimates say? The bottom panel of the historical decomposition shown in my original guest blog indicates that their explanation does not line up particularly well with the data. For example, the component of the real price of oil explained by the hybrid speculative shock is essentially flat during the third quarter of 2008 and flat again starting in early 2009. In fact, the cumulative effect of the hybrid speculative shock declined in the last quarter of 2008. All of this happened, while according to Juvenal and Petrella the market expected higher spot prices. This example shows that contango in the oil futures market does not imply hybrid speculation. The corresponding evidence for the first contango episode is just as disappointing. The component of the real price of oil explained by the hybrid shock is essentially flat until mid-2005 and again after mid-2006. Does this evidence not contradict the explanation proposed by the authors?
Not only is this additional evidence not persuasive, but I do not see what this evidence has to say about the central question of whether financial speculation caused the surge in the real price of oil, given the difficulty of attributing changes in the term structure to exogenous decisions by index funds and other financial investors. To conclude, nothing Juvenal and Petrella added in their rejoinder overturns my initial concerns. I have yet to see solid evidence for the claim that financial speculation played a significant role in causing the run-up in the real price of oil.