Navigating the Oil Bubble: A Non-linear Heterogeneous-agent Dynamic Model of Futures Oil Pricing.

AuthorCifarelli, Giulio

    Between 2003 and 2019, oil prices witness two major cycles, which straddle the 2009 global financial freeze. From 2003 to 2009, oil prices experience an extremely sharp upswing and subsequent collapse in the last two years, variously attributed to shifting fundamentals, institutional changes and/or to financial bubble behavior. This gave rise to an extensive literature that investigates the role of speculators in the oil futures market. A selective survey of this literature provides the following indications.

    On the one hand, Hamilton (2009) and Kesicki (2010), relate oil price fuctuations observed between 2003 and 2009 to changes in fundamental variables (weak dollar combined with low elasticity of supply). In a similar vein, Kilian and Murphy (2014) and Juvenal and Petrella (2015) conclude that oil price dynamics is related to fundamentals, speculation playing but a minor role. Fattouh et al. (2013) reach similar conclusions in their extensive survey on the literature that investigates role of speculation in the oil markets after 2003. On the other hand, Master (2008) and Sari et al. (2012), among others, attribute the upswing in oil prices to the long positions of institutional investors entering the commodity markets in the context of financialization of commodity markets (on this see also Tang and Xiong 2012). More recently, financial herd behavior has emerged as a possible interpretation of this phenomenon, expanding the literature on financial determinants in commodity pricing, which Demirer et al. (2015) and Boyd et al. (2018) summarize.

    With the world economy plunged in the Great Recession and with major technological innovations (shale oil in particular) and geopolitical turmoil (Middle-East conficts, Saudi Arabia energy policy shifts) affecting the global oil industry, a proper identification of oil price drivers, in the final ten years of the sample (2010 -2019), becomes more dificult. (1) Indeed the market witnesses confusing changes in demand and supply factors, leaving it open to question, whether a financial interpretation of oil price swings maintains its relevance. (2)

    Among the studies that emphasize the role of the financial determinants of oil prices, one strand explains the 2007-2009 price upswing as the result of a speculative bubble. Empirical evidence in this respect is provided by Shi and Arora (2012) and Brooks et al. (2015) using the SADF explosive behavior econometric test of Phillips et al. (2011). A common weakness of their approaches--as pointed out by Gronwald (2016)--is their reliance on a measure of fundamentals, which is likely to be arbitrary. A recent strand of the bubble literature, based on the so-called log-periodic power law (LPPL) model set out by Sornette and Johansen (1997) and applied to oil pricing by Zhang and Yao (2016), bypasses this problem by focusing on the peculiarities of the bubble dynamics. According to this approach, the traders' purchase and sale decisions are not taken in isolation. They are influenced by those of other traders and by the external environment. As a result, as pointed out by Geraskin and Fantazzini (2013), there arises a nonlinear positive feedback reaction which results in unsustainable dynamic processes, defined bubbles, in which supply and demand considerations become insignificant. Imitation sparks self-reinforced runs, which result in super exponential price growth, whereas self-similar herding causes log periodic price oscillations (Zhou and Sornette, 2006). (3)

    In this paper we focus on the analysis of the 2007-2009 upswing, seen as a highly informative case study of a bubble-like phenomenon, the empirical research being motivated by the desire to associate some major intuitions provided by the LPPL bubble approach with the tenets of a model based on the Heterogeneous Agents Model (HAM) literature. (4) The value added of this approach is to reformulate the novel insights emerging from the LPPL literature in terms of a complete and fexible model of the oil market, where different categories of agents interact and influence each other. Our approach makes it possible to identify the origin of the destabilization, associating it with some criteria (e.g. imitation) derived from the LPPL literature.

    Our modelling strategy draws inspiration from Sornette et al. (2009) and their painstaking identification of LPPL bubble-dynamics in oil (futures) prices over 2007-2009. The latter is generated by speculative behavior attributed--in a context of growing uncertainty on global oil demand and supply--to the interplay of various factors, such as protective hedging against future oil price increases, search for higher prospective (financial) portfolio investment yields in the oil futures sector and deregulation in 2006 of oil futures trading by the US Commodity Futures Trading Commission. These hypotheses are incorporated in our model and explicitly tested for.

    Expanding on Westerhof and Reitz (2005), Reitz and Westerhof (2007) and Tokic (2011) among many others, our model attributes the oil bubble to alterations in the standard reaction of market participants, feedback traders, fundamentalist speculators and hedgers, due to the oil price vagaries mentioned above.

    More precisely, we set no a priori restrictions on the signs of the parameters of the futures return relationship. In the same way, no restrictions are imposed on the sign of the speed of adjustment coefficient in the logistic functions, which model the entry in (exit from) the market of agents according to their trust in the reliability of market pricing. In order to account for the time span of the bubble a dummy is introduced in the nonlinear functional relationships. The hypothesis is that these modifications in the standard reaction of market participants will explain the bubble-like price dynamics identified by the LPPL analysis.

    We also introduce two indicators to control for currency and financial market conditions, finding that changes in weighted US dollar exchange rate and the VIX (VOX) index have a statistically significant impact on oil pricing patterns.

    Based on the Zhou-Sornette (2009) and on the Phillips et al. (2011) methodologies, we identify the presence of a single bubble between January 2007 and February 2009. The bubble affects all categories of agents, in some cases reinforcing their behavior (as observed over the whole sample period) in other cases altering it. More specifically, the bubble tends to bring about a stabilizing reaction from hedgers and chartists (acting as contrarians) and to reinforce the market destabilizing behavior of fundamentalists. These results apply to the entire sample (2003--2019) and are strongly corroborated over the 2003-2009 subset.

    By modelling both the one-month and the three-month to expiry futures contracts, based on weekly data, we test whether contract maturity affects these patterns. (5) As expected, in periods of turmoil and rising uncertainty, such as the 2003-2009 time interval, rational agents react to an increase in perceived disequilibrium by shifting from the one-month to the relatively less volatile three-month contract.

    The main contributions of our analysis can be summarized as follows.

    First, the HAM model has a persistent explanatory power, in spite of the extreme turbulence of the time period under investigation. Oil price dynamics is driven by a set of financial factors, mostly related to the behavior of feedback traders, fundamentalist speculators and hedgers. Shifts in oil excess demand have an impact on prices only insofar as they affect the reaction of these agents. In the same way, the stock exchange expected volatility and US dollar trade weighted changes are not the main drivers of oil futures price hikes. The role of fundamental variables tends to be mostly indirect.

    Second, the empirical results suggest that the flexibility of the HAM approach can be substantially increased, as shown, among others, by Tramontana et al. (2010). No a priori restriction must be imposed on the signs of the nonlinear behavioral relationships of the model, least we introduce misspecifications. In periods of uncertainty standard reactions cannot be discounted, and have to be ascertained empirically.

    Third, introducing slope dummies in (the nonlinear reaction functions of) a heterogeneous agent model, it is possible to capture alterations in the behavior of speculators and hedgers that are specifically related to the bubble detected by the LPPL analysis and provide relevant insight as to its origin. We identify the increasingly destabilizing behavior of fundamentalist speculators as the single most relevant factor driving the 2007-2009 oil price bubble. These results are in line with Zhang and Wu (2019) who find that hedge funds play an important role in pushing up crude oil price during the 2008 bubble, their impact being stronger when crude oil futures prices rise substantially.

    Fourth, in view of the nature of the HAM model and the relative freedom it allows in structuring our components, we have conducted systematic robustness tests that support our preferred specification. In detail, our tests corroborate the selection of three categories of agents, chartists, fundamental speculators and hedgers, and the choice of a LSTAR parameterization of their nonlinear reaction functions.

    This research is structured as follows. Section 2 analyses the theoretical and empirical characteristics of our three-agent model. Section 3 sets forth the empirical estimates over the periods. Section 4 concludes the paper and provides an economic and financial interpretation of the observed oil futures price gyrations.


    2.1 Theoretical considerations

    As Manera (2013) recalls, speculation can be defined as the activity of buying or selling in futures markets in the expectation of future price movements to make profits, as opposed to hedging, when people buy or sell with the ultimate aim of taking delivery...

To continue reading

Request your trial

VLEX uses login cookies to provide you with a better browsing experience. If you click on 'Accept' or continue browsing this site we consider that you accept our cookie policy. ACCEPT