Home > Research > Publications & Outputs > Using Market Expectations to Test for Speculati...

Electronic data

  • Pavl_Paya_Peel_MS17_337

    Accepted author manuscript, 469 KB, PDF document

    Embargo ends: 3/07/20

    Available under license: CC BY-NC: Creative Commons Attribution-NonCommercial 4.0 International License

Links

Text available via DOI:

View graph of relations

Using Market Expectations to Test for Speculative Bubbles in the Crude Oil Market

Research output: Contribution to journalJournal article

Published
<mark>Journal publication date</mark>1/08/2018
<mark>Journal</mark>Journal of Money, Credit and Banking
Issue number5
Volume50
Number of pages24
Pages (from-to)833-856
Publication statusPublished
Original languageEnglish

Abstract

The wide fluctuations of oil prices from 2003 to 2008 have attracted the interest of academics and policymakers. A popular view is that these fluctuations were caused by speculative bubbles due to the increased financialization of oil futures markets. This hypothesis, however, is difficult to examine since the fundamental price of oil is unobservable and, therefore, econometric evidence in favor of bubbles may actually be due to misspecified market fundamentals. In this paper, we extend two recently proposed methodologies for bubble detection that alleviate this problem by using market expectations of future prices. Both methodologies provide no evidence of speculative bubbles.

Oil price shocks may be associated with significant changes in the overall economic performance of countries (Kilian 2008a, Hamilton 2009, Blanchard and Galí 2010, Hamilton 2013, Kilian 2014, Baumeister and Kilian 2016c). As such, they have been subject to detailed scrutiny for more than three decades (for comprehensive reviews, see Hamilton 2013 and Baumeister and Kilian 2016a). Traditionally, research about the key drivers of oil price fluctuations has been dominated by arguments about fundamental supply and demand shocks (Barsky and Kilian 2002, Hamilton 2003, Kilian 2008b, 2009). It is only recently that the role of speculation has been put forward as a serious alternative (see, e.g., Hamilton 2009, Kilian and Murphy 2014, Knittel and Pindyck 2016).

One of the reasons for this recent emphasis on speculation is the eruption of index trading and financialization of oil markets from 2004, and an almost simultaneous surge in the price of oil. The focus on speculative pressures was further fueled by press coverage of the testimonies of Michael Masters before the U.S. Senate (Masters 2008, 2010). The possibility that financial speculation plays a central role in the determination of oil prices naturally raises the issue of whether bubbles are a feature of the dynamics of oil prices. This view has found seeming support by studies providing evidence of bubbles in oil prices in the 2000s (see, e.g., Phillips and Yu 2011, Lammerding et al. 2013, Tsvetanov, Coakley, and Kellard 2016).

Studies based on structural models, on the other hand, provide strong evidence that the surge in the real price of oil was driven primarily by shifts in the demand for oil associated with global business cycle fluctuations and that speculation did not have an important impact on the spot price (for a summary of this literature, see Fattouh, Kilian, and Mahadeva 2013). For example, Kilian and Murphy (2014) develop a structural vector autoregressive model of the global market for crude oil that allows for shocks to the speculative demand for oil. They find that expectations of agents about future demand and supply conditions not already captured by flow demand and flow supply shocks accounted for less than 10% of the long‐run variation in the real price of oil. The conclusion that speculative demand did not play a significant role in the determination of the real price of oil since 2003 is robust to the use of alternative proxies for inventory data and a wide range of values of price elasticities of oil demand (Kilian and Lee 2014).

In this paper, we contribute to this debate by employing prices and market expectations to analyze the presence of speculative bubbles in the crude oil market. In particular, we extend two recent methodologies, proposed by Pavlidis, Paya, and Peel (2017), to test for periodically collapsing bubbles. The key idea behind these methodologies is simple: in the presence of an ongoing speculative bubble, future spot and expected prices will diverge because market participants rationally attach a nonzero probability to the bubble bursting when forming expectations. Under general conditions, the deviation of the future spot price from the expected price will be a function of the bubble process (and, as such, explosive) but it will not depend on market fundamentals. Consequently, one can simply test for bubbles, first, by running recursive right‐tailed unit root tests on the difference between future spot and expected prices and, second, by sequentially testing the unbiasedness hypothesis in the oil market.

Various unit root tests have been proposed in the literature to test for explosive dynamics in time series (see, e.g., Gürkaynak 2008, Homm and Breitung 2012). In principle, any of these could be used to investigate the existence of bubbles. We choose to work with the Generalized Supremum Augmented Dickey Fuller (GSADF) of Phillips, Shi, and Yu (2015a, 2015b). Simulation evidence in Phillips, Shi, and Yu and Pavlidis, Paya, and Peel (2017) suggests that the GSADF displays good size properties and is superior to its rivals in detecting multiple bubble episodes. Furthermore, this test is accompanied by a date‐stamping strategy that permits the identification of the exact periods of market exuberance. With regard to testing the unbiasedness hypothesis, we adopt the rolling‐window approach of Pavlidis, Paya, and Peel (2017). This approach consists of sequentially estimating predictive regressions and drawing statistical inference using the IVX instrumentation method of Phillips and Magdalinos (2009), Phillips and Lee (2013), and Kostakis, Magdalinos, and Stamatogiannis (2015). The IVX method is particularly attractive in this setting because it allows robust chi‐square inference for a wide range of autoregressive processes, from stationary to mildly explosive.

The main advantage of the above methodologies is that, by exploiting the information incorporated in market expectations, they do not require the specification of market fundamentals and, thus, ameliorate the well‐known joint hypothesis problem. Both methodologies require, instead, a proper measure of market expectations.1 Pavlidis, Paya, and Peel (2017) use derivative prices to proxy for expectations in the foreign exchange and stock markets. However, derivative prices are contaminated by a risk premium, which, if large and persistent, may confound the analysis. In this paper, we employ two alternative measures of expectations that do not suffer from this shortcoming. The first measure is obtained as in Baumeister and Kilian (2016b) by adjusting futures prices by an estimate of the risk premium based on the term structure model of Hamilton and Wu (2014). The second is directly obtained from survey data. By applying the GSADF and IVX tests to price and expectation data on the crude oil market from 1990 to 2013, we find that the two expectation measures yield qualitatively similar results. Overall, there is no evidence in favor of speculative bubbles. This is despite the fact that oil prices and expectations individually exhibit periods of explosive dynamics. Thus, our findings are consistent with the literature that supports the view that oil price movements are driven by changes in fundamental factors.

The rest of the paper is structured as follows. Section 1 describes the theoretical framework. Section 2 discusses the evidence of risk premia in the oil futures market and the implications for tests based on derivative prices. Section 3 deals with the measurement of market expectations. Section 4 provides a brief outline of the econometric methods employed in the paper. Section 5 presents the empirical application, and the final section concludes.