Energy Economics xxx (2015) xxx–xxx

ENEECO-03051; No of Pages 6

Contents lists available at ScienceDirect

Energy Economics

j ourna l homepage: www.e lsev ie r .com/ locate /eneco

Date stamping historical periods of oil price explosivity: 1876–2014☆

Itamar Caspi a,b, Nico Katzke c,⁎, Rangan Gupta d

a Bank of Israel, Israelb Bar-Ilan University, Israelc Department of Economics, Stellenbosch University, South Africad Department of Economics, University of Pretoria, Pretoria 0002, South Africa

☆ The views expressed in this paper are those of the areflect that of the Bank of Israel.⁎ Corresponding author.

E-mail addresses: itamar.caspi@boi.org.il (I. Caspi), nicrangan.gupta@up.ac.za (R. Gupta).

1 C.f. Gürkaynak (2008) for an in-depth assessment of tble detection techniques.

http://dx.doi.org/10.1016/j.eneco.2015.03.0290140-9883/© 2015 Elsevier B.V. All rights reserved.

Please cite this article as: Caspi, I., et al., Dadx.doi.org/10.1016/j.eneco.2015.03.029

a b s t r a c t

a r t i c l e i n f o

Article history:Received 11 September 2014Received in revised form 25 February 2015Accepted 19 March 2015Available online xxxx

JEL classification:C15C22

Keywords:Oil-pricesDate-stamping strategyPeriodically collapsing bubblesExplosivityFlexible windowGSADF testCommodity price bubbles

This paper sets out to date-stamp periods of historic oil price explosivity using the Generalized sup ADF (GSADF)test procedure developed by Phillips, Shi, and Yu (2013). The date-stamping procedure used in this paper is ef-fective at identifying periodically collapsing bubbles; a feature found lacking with previous bubble detectionmethods.We set out to identify periods of oil price explosivity relative to the general price level and oil inventorysupplies in the US since 1876 and 1920, respectively. The recursive identification algorithms used in this studyidentify multiple periods of price explosivity, and as such provides future researchers with a reference for study-ing the macroeconomic impact of historical periods of significant oil price build-ups.

© 2015 Elsevier B.V. All rights reserved.

1. Introduction

The occurrence of bubble episodes in asset markets have been stud-ied extensively, both theoretically and empirically, and have spurred di-vergent debates on its implications on rationality andmarket efficiency.Economists have long debated whether to reconcile bubble-like behav-ior with rational expectations of future prices, leading to divergentviews on suitable policy responses following its detection. To this end,various econometric techniques have been proposed for date stampingpast bubble periods as well as suggesting mechanisms for early detec-tion of its formation.1

This paper sets out to identify periods ofmildly explosive behavior ofthe oil price during the period January 1876 to January 2014. We do sousing Phillips, Shi, and Yu's (2013) recursive Generalized sup Augment-ed Dickey–Fuller (GSADF) technique applied to test for significant devi-ations of the nominal price of oil from its prior levels relative to the

uthors, and do not necessarily

okatzke@sun.ac.za (N. Katzke),

he performance of various bub-

te stamping historical perio

general price level in the US and the US inventory supply stock. Thistechnique allows the detection and date-stamping of periods wherethe price behavior, relative to general prices and oil supply, resemblesan explosive series.2 In this study we define such explosivity as tempo-rary regime shifts from unit-root non-stationarity toward periodswhere the root significantly exceeds unity, followed by reversion backto amartingale process. After date-stamping,we contextualize the iden-tified periods of explosiveness in terms of contemporaneous events thatmay have contributed to the said price surges.

Most applications of the techniques developed by Phillips et al.(2013) and used in this study, interpret such defined periods of explo-siveness as bubble-periods. An asset price bubble can be defined theo-retically as sustained price deviations from an identified fundamentalvalue.We, however, apply caution as to the interpretation of our results,defining such periods as explosive as opposed to indicative of an oilprice bubble.

Our caution in defining explosivity measured in this paper as bub-bles follows for several reasons. Firstly, defining a fundamental levelproves uniquely complicated for storable commodities. Similar studiesinto the explosiveness of commodity prices have typically made use of

2 Phillips and Magdalinos (2007) elaborate on the limit theory of mildly explosive be-havior, referred to in this paper merely as explosivity.

ds of oil price explosivity: 1876–2014, Energy Econ. (2015), http://

Table 1Real oil price explosivity.

Sample: 1876M01 2014M03

Included observations: 1659

Starting date Ending date Duration (months)

1895M01 1895M05 51899M11 1900M03 41909M11 1911M01 151912M11 1914M02 161946M05 1948M10 301951M01 1952M10 221964M07 1964M11 51966M02 1966M07 61970M06 1970M11 61973M10 1977M03 411979M04 1982M03 361986M02 1986M07 62007M10 2008M08 11

Table 2Nominal price and inventory quantity ratio explosivity.

Sample: 1920M01 2014M02

Included observations: 1129

Starting date Ending date Duration (months)

1949M02 1949M06 51973M09 1984M12 1361985M07 1985M12 62005M07 2005M10 42006M04 2006M08 52007M07 2008M09 15

2 I. Caspi et al. / Energy Economics xxx (2015) xxx–xxx

Pindyck's (1993) convenience-yield in order to define the fundamentalprice of, e.g., oil. This ismeasured as the sumof discounted oil “dividends”,which is in turn approximated for by the benefit to holding inventoriesper unit of commodity over a defined period to which a futures contracthas beenwritten on the underlying asset.3 This can be thought of as anal-ogous to the dividend earned on holding a share. In this study, as we lacklonger dated futures contract data,we define the underlying fundamentallevels of the oil price as relative to the general price level and the inven-tory supply level in the US (supply ratio hereafter).

We also apply caution to labeling the identified periods ofexplosivity as bubbles, as such periods might rather be indicative ofadjustments from previously managed or manipulated pricing sched-ules toward a more fundamental level (whichever way defined). Toinfer exuberance, or bubble-like behavior, would require the implicit as-sumption that the pricewas at its fundamental level prior to explosivity.This is clearly not plausible in the stronglymanipulated oil pricemarketfor a commodity with a particularly hard to define fundamental level.Our interpretation thus of the oil price series is that it is non-linearly re-lated to underlying fundamentals.

Despite our caution on interpretation, our results should still proveuseful. This follows as the price of oil and its derivatives are an impor-tant input into nearly all spheres of modern economies. Regnier(2007), e.g., also shows that production prices for commodities closelytrack the oil-price, which would imply a significant impact on generalprice stability in the economy should the latter experience periods ofexplosivity. While commoditymarkets, and in particular the oil market,have well functioning derivative markets that allow effective hedging,large price fluctuations remain costly to households in the economywho participate chiefly in consuming oil derivative products in theform of fuel and gas. By identifying significant build-ups in the price ofoil relative to other prices in the economy and relative to the supply,our results provide researcherswith a reference to assess themacroeco-nomic impact of historical periods of oil price explosivity.

The paper is structured as follows: Section 2 discusses the relevantliterature and theory of asset pricing explosivity, while Section 4 pre-sents thedata used in this study. Section 3 discusses the empiricalmeth-odology followed, while Section 5 discusses the results. The datesidentified by our GSADF approach as representing periods of oil priceexplosivity are summarized in Tables 1 and 2.

4 C.f. Branch and Evans (2011) for a detailed description of the rational price bubble lit-erature; Gürkaynak (2008) also provides a thorough account of the broad literature onempirical tests for bubbles.

2. Literature overview

The inflationary build-up of asset prices, more loosely defined asasset price bubbles, has long interested economists and led to the

3 C.f. Lammerding et al. (2013) for a deeper discussion into this definition.

Please cite this article as: Caspi, I., et al., Date stamping historical periodx.doi.org/10.1016/j.eneco.2015.03.029

development of a vast literature aimed at explaining its existence andfacilitating its timely detection and prevention. The difficulty in testingfor the presence of bubbles lies in modeling its explosivity and labelingits occurrence. Traditional unit root and co-integration tests aimed atidentifying such periods, as e.g. proposed by Diba and Grossman(1988), may not bear out the existence of bubbles when they are peri-odically collapsing.4 As Evans (1991) points out, when seeking to iden-tify multiple periodically collapsing bubbles within a single data setusing stationarity tests, the process is greatly complicated and exposedto the possibility of identifying pseudo stationary behavior.

To overcome this problem, Phillips and Yu (2011), Phillips, Wu, andYu (2011) and Phillips, Shi, and Yu (2013) (PY, PWY and PSY, respec-tively, hereafter) developed and subsequently improved on a convinc-ing sequence of rolling right-tailed sup ADF testing procedures todetect and date stamp mildly explosive pricing behavior. Hommand Breitung (2012) compared severalwidely used techniques for iden-tifying bubbles and found that the PWY (2011) strategy performs thebest. PSY (2013) extended the methodologies of PWY (2011) and PY(2011) in that it recursively identifies explosivity as rejecting the nullhypothesis of unit-root non-stationarity for the right-tailed alternativeof explosivity. They found this strategy to significantly outperform pre-viously used right-tailed ADF estimations in identifying multiple bub-bles using Monte Carlo simulations. In particular, the PSY (2013)approach overcomes the earlier mentioned problem of detectingmulti-ple episodes of periodically recurring explosivity (a particularly usefulfeature, as in our study we identify several), and has since gainedground in its empirical applications (c.f. inter alia Bettendorf and Chen(2013), and Etienne et al. (2014)).

Most studies focussing on commodity price explosivity have soughtto identify such periods using Pindyck's (1993) convenience yield (c.f.inter alia Lammerding et al. (2013), Gilbert (2010); Areal et al. (2014)and Shi and Arora (2012)). This cost-of-carry equation is then used toapproximate the fundamental value of the oil price,5 to which PSY's(2013) estimation procedures have been applied to identify significantdeviations from it. Lammerding et al. (2013) separate the fundamentallevel from the unobserved “bubble” component by expressing the stan-dard present-value model of discounted future oil dividends in statespace form. They then approximate two distinct Markov Switchingphases to distinguish between the stable and explosive phases of thebubble process. Their approach uncovers robust evidence of speculativebubbles present in oil price dynamics. Shi and Arora (2012) apply threedifferent regime switching bubble-identifying procedures, finding evi-dence of a short-lived real oil price bubble between 2008 and 2009.

In these studies the authors use daily futures prices from contractstraded on the New York Mercantile Exchange (NYMEX) for West

5 Intuitively, it approximates the fundamental value of oil from the current and expect-ed discounted convenience yield that accrues from holding oil inventories.

ds of oil price explosivity: 1876–2014, Energy Econ. (2015), http://

3I. Caspi et al. / Energy Economics xxx (2015) xxx–xxx

Texas Intermediate (WTI6) prices, and Inter-Continental Exchange (ICE)Futures Europe for Brent prices. The starting points are all after 1983 forWTI and 1989 for Brent crude prices due to data availability on futurescontracts. As our study is based on prices dating back to 1876,we cannotuse similar futures data to approximate a convenience yield. Our ap-proach is then to use the US general price series data and oil inventoriesin the US as relative series to identify periods of unsustainable oil pricebuild-ups. This is discussed in more detail in Section 3.

3. Theoretical framework and methodology

The techniques used in this study build upon the work pioneered byPhillips and Yu (2011) and Phillips, Wu, and Yu (2011), and specificallythe generalized form of the SADF (GSADF) technique as suggested byPhillips, Shi, and Yu (2013). The latter was designed to overcome thePY and PWY procedure lack of power in identifying a second bubble ifit is dominated by thefirst. To do so, PSY develop in their paper aflexiblemoving sample test procedure that is able to consistently detect anddate-stamp multiple bubble episodes and seldom give false alarms,even in modest sample sizes. The PSY approach achieves this by recur-sively implementing an ADF-type regression test using a rolling win-dow procedure. More specifically, we consider an ADF regression for arolling interval beginning with a fraction r1 and ending with a fractionr2 of the total number of observations, with the size of the windowbeing rw = r2 − r1.

Let:

yt ¼ μ þ δ:yt−1 þXp

i¼1

ϕirwδ:yt−i þ ϵt ð1Þ

where μ, δ and ϕ are parameters estimated using OLS, and the usualH0 : δ = 1 then tested against the right sided alternative H1 : δ N 1.7

The number of observations taken into account by 1 is Tw = [rwT],where [.] is the integer part. The ADF statistic corresponding to 1 is de-noted by ADFr2r1 .

PSY (2013) formulated a backward sup ADF testwhere the endpointof the subsample is fixed at a fraction r2 of the whole sample and thewindow size is expanded from an initial fraction r0 to r2. The backwardsup ADF statistic is then defined as:

SADFr2 r0ð Þ ¼ supr1∈ 0;r2−r0½ �ADFr2r1 : ð2Þ

The generalized sup ADF (GSADF) is then constructed by repeatedlyimplementing the SADF test procedure for each r2 ∈ [r0, 1]. The GSADFcan then be written as follows:

GSADF r0ð Þ ¼ supr2∈ r0 ;1½ �SADFr2 r0ð Þ: ð3Þ

The rationale behind using a supremum of a recursively estimatedADF statistic is the observation that asset price bubbles generally col-lapse periodically, with conventional unit root tests then having limitedpower in detecting such bubbles (Evans, 1991). Homm and Breitung(2012) argued that the sup ADF test procedure delivers a fairly efficientbubble-detection technique when dealing with one or two bubbleperiods.

This process can be understood as a recursive regression process cal-culated by Eq. (1) using OLS with the initial fraction rw = r0, and then

6 WTI (also known as Texas Sweet Light) is typically regarded as the reference oil typefor the US.

7 As mentioned by an anonymous reviewer, this process could also be re-estimatedusing a fractionally integrated mean process. This would entail estimating a right-tailedADF test, with δ→ 1 corresponding to a unit root, while δ N 1 corresponds to explosivity.The benefit of this approach is that it incorporates long-memory into the process, andmayyield more accurate results. This, however, remains an avenue for future methodologicalresearch using our approach. C.f. Cuñado et al. (2007) for the use of fractionally integratedestimation procedures to detect bubbles.

Please cite this article as: Caspi, I., et al., Date stamping historical periodx.doi.org/10.1016/j.eneco.2015.03.029

expanding the sample window forward until rw = r1 = 1, which isthe full sample. The initial minimum fraction is selected arbitrarilywhile keeping in mind that we need to ensure estimation efficiency.This process is then repeated for each possible fraction, and an ADF sta-tistic calculated as ADFrk for all values of k ∈ (r0, r1). This procedure re-sults in a sequence of ADF statistics. The supremum value of thissequence (SADF) can then be used to test the null hypotheses of unitroot against its right-tailed (mildly explosive) alternative by comparingit to its corresponding critical values. Significant ADF statistics are indi-cated by δr1 ;r2N1, which we could then label as explosive (bubble) pe-riods. The GSADF approach defined above uses a variable windowwidth approach, allowing starting as well as ending points to changewithin a predefined range [r0, 1], and thereby allowing the identificationof several bubble periods consistently in a sample.

The procedure described above is used to test whether a certain timeseries exhibits explosive patterns, interpreted as bubbles, within a givensample. If the null hypothesis of no bubbles is rejected, the PSY proce-dure enables us, as a second step, to consistently date-stamp the startingand ending points of this (these) bubble(s). The starting point of a bub-ble is defined as the date, denoted asTre (in fraction terms), atwhich thebackward sup ADF sequence crosses the corresponding critical valuefrom below. Similarly, the ending point of a bubble is defined as thedate, denoted as Tr f (in fraction terms), at which the backward supADF sequence crosses the corresponding critical value from above.

Formally, we can define the bubble periods based on the GSADF testby8

r̂e ¼ infr2∈ r0 ;1½ �

r2 : BSADFr2 N cvβTr2

n o

r̂ f ¼ infr2∈ r̂e ;1½ �

r2 : BSADFr2 N cvβTr2

n o ð4Þ

where cvβTr2

is the 100(1 − βt)% critical value of the sup ADF statistic

based on Tr2

h iobservations.We also setβt to a constant value, 5%, as op-

posed to letting βT → 0 as T → 0. The GSADF (r0) for r2 ∈ [r0, 1] is thebackward sup ADF statistic that relates to the GSADF statistic by notingthat:

GSADF r0ð Þ ¼ supr2∈ r0 ;1½ �

BSADFr2 r0ð Þn o

: ð5Þ

We face several challenges dating historic oil price bubbles. Firstly,as we do not have sufficiently long dated forward price data, we cannotrely on estimates of the convenience yield for a fundamental level, asdiscussed in Section 2. As such, we base our analysis on two measures:firstly that of defining periods of explosivity in the oil price relative tothe general price level in the economy. We thus implicitly assume thatthe value of oil should keep some parity relative to other goods in theeconomy, at least in the short term. The second approach considersthe nominal price of oil relative to the inventories held in the US, usedas a proxy for the available supply of oil. This way we test for significantoil price deviations per unit of oil inventory in the US, which is moreclosely related to work done by Reitz and Slopek (2009), where the au-thors calculate a fundamental value of the oil price by assuming that itdepends on excess capacity in oil production.

Another challenge faced in our estimation is dealing with the longperiod of virtual oil price stagnation between 1941 and 1973. Identify-ing periods of explosivity during this period for either of our measureswould likely constitute general price inflation build-ups or oil supplyshocks, as opposed to price fluctuations. As such, we caution the valueof identifying episodes of explosivity during this period.

8 For a more in-depth discussion of this process of bubble identification, refer to PSY(2013).

ds of oil price explosivity: 1876–2014, Energy Econ. (2015), http://

0

10

20

30

40

50

60

70

0

40

80

120

160

200

240

280

1890 1910 1930 1950 1970 1990 2010

Log(Real Oi l Price) US CPI

Fig. 1. Logarithm of real oil price: 1920–2014.

4 I. Caspi et al. / Energy Economics xxx (2015) xxx–xxx

4. Data description

This study uses themonthlyWTI crude oil price series for the period1876–2014, obtained from the Global Financial database.9 US CPI data,also obtained from the Global Financial database, is then used to calcu-late the real oil price for this period. Similar to PY (2011) we approxi-mate oil-supply from the US inventory of crude oil.10 The inventorydata is obtained from theUSDepartment of Energy and is used through-out this paper as a proxy for the oil supply in the economywith the larg-est consumption of oil.

Fig. 1 plots the logarithmic scale of the deflated price of oil for the pe-riod 1876–2014, as well as the US CPI series. The rescaling of the realprice controls for level distortions by approximating the relative chang-es in prices. From the figure it is clear that the real oil price has shownconsiderable variation both before and after the largely stagnant priceperiod between 1941 and 1973.11 Over the two subsamples, 1876–1940 and 1973–2014, the real price of oil averaged respectively $16.78and $23.97. The period after 1973 was much more volatile (with stan-dard deviations after the stagnant price period double what it was be-fore), with oil cartel influence and supply shocks causing greatfluctuations in the real price of oil. As can be seen too from the figure, ag-gregate prices in the US rose steadily after the 1970s, displaying a gen-erally stable upward trend. This is important, as our interpretation ofreal oil price explosivity would have been distorted if, e.g., there wereperiods of significant deflation and oil prices remained stagnant.12

Next we consider Fig. 2, depicting the logarithmic transform of thenominal price supply ratio, aswell as the level quantity of oil inventoriesin the US (in 1000s) since 1920. From the figure it follows that the ratiodeclines sharply after 1920, largely as a result of a sharp increase in theUSoil supply. Thereafterwe see that the supply ratio remains largelyflatuntil mid 1973, after which the price of oil rises significantly. Thereafterwe see several periods of significant build-ups in the ratio series. Duringthis period, oil supply initially also rises, but thereafter remains largelyflat without any significant periods of sustained decline. This again isimportant as periods of decline might similarly as before distort our in-terpretation of oil price explosivity.

Therefore, aswe do not find any periods of significant explosivity foreither the inventory supply or the inflation series that correspond to areduction in the respective series which are contemporaneous withour identified periods of oil price explosivity, we view such periods ofexplosivity as indicative of significant oil price increases.13

5. Empirical results

In order to motivate the use of our GSADF analysis, we first conductunit root tests on both the real price and the price ratio series for the fullsample and sub-sample periods. The Augmented Dickey–Fuller (ADF)test suggests that for both series the unit root hypothesis cannot berejected for the full sample, as shown in Table B.4. However, for thesub-samples between 1876–1941 and 1920–1973 for the two series re-spectively, we reject the hypothesis. This result may, however, be a re-sult of the ADF test being biased in the presence of structural breaks.Considering Table B.5, where we control for breaks in each series forboth the intercept and the trend using the Zivot–Andrews test, wefind that the two sub-samples are indeed I(1), but only at the 1% level.We then employed both the Lumsdaine and Papell (1997) and the

9 Listed as WTI—Cushing, Oklahoma.10 Department of Energy, Monthly Energy Review, series: US Crude Oil ending stocksnon-SPR (thousands of barrels).11 In December 1941 the Wartime Price and Trade Board announced that a completecontrol of all oil priceswould come into force. This controlwas subsequently lifted in 1973.12 This follows as CPI is the denominator in the real price ratio, and such a scenario couldimply labeling periods of explosivity despite the nominal price of oil potentially remainingflat.13 Results for the CPI and inventory supply GSADF tests are omitted for the sake of brev-ity, but can be requested from the authors.

Please cite this article as: Caspi, I., et al., Date stamping historical periodx.doi.org/10.1016/j.eneco.2015.03.029

minimum-LM Lee and Strazicich (2003) tests, which both allow fortwo endogenous breaks in the respective series, in order to establishwhether there is a unit root in the data for both these sub-periods.14

The results are reported in Tables B.6 and B.7, and show that indeedthe series are I(1) for both sub-samples when we account for multiplebreaks in the data. These findings are consistent with Hamilton's(2009, p.3) discussion on the unpredictable unit root behavior of theprice of oil.

Our results for the right tailed ADF tests for explosivity are summa-rized in Table B.3 in the appendix.15 The output suggests significantright tail deviations from unit root-behavior for both series and foreach of the sub-samples, except for the ratio series during 1920–1941.Explosivity identified for this period should thus be interpreted withcaution.

Next, we set out to date-stamp the periods of price explosivity of theoil price relative to the general price level and US inventory supplies.The periods of real oil price and price supply ratio explosivity identifiedusing the GSADF approach of PSY (2013) are summarized in Tables 1and 2 below.

The real oil price shows, prior to the stagnant price period, severalepisodes of real oil price build-up. The first price spike identified in1895 coincides with the loss of oil production in the Appalachian fieldsand a cholera epidemic in Baku in 1894 (Hamilton, 2011, p. 4). Theslight oil price spike in 1900 could also likely be attributed to the docu-mented surge in demand for the new uses of oil apart from fabricatingilluminants at the time.16

Following the Great Depression in 1929, oil demand declined sharp-ly, coupledwith large oil discoveries at the time. The nominal price of oilslid dramatically following the Depression in 1931, plummeting over160% from its peak in April 1929 of $24.2 to $9.19 in July 1931. Pricesthen rose moderately thereafter as output continued a slide downwardfrom its peak in 1930.

The period after 1940 saw global oil prices and supply remainedlargely stagnant, despite accelerating demand following World War II.The supply ratio GSADF results suggest two periods of explosivity, al-though this was entirely production driven and as such not a price phe-nomenon. Nominal oil prices did rise somewhat during 1945–1947,while significant oil disruptions followed in 1952–1953,whichwas con-temporaneous with the Korean conflict and Iranian efforts to national-ize their oil industry. While these events, coupled with the Suez Crisisof 1956–1957, caused some global concern on oil markets and despitethe general unrest and uncertainty, the quantity of oil produced didnot change much during this period. During the late 1960s, modest

14 For the sake of brevity, we omit a deeper discussion into these tests.15 Estimations were carried out using Caspi's (2013) add-in in Eviews 8.16 E.g. Petroleum gained in use for commercial and industrial heating, as well as for railand motor vehicle transportation (Hamilton, 2011).

ds of oil price explosivity: 1876–2014, Energy Econ. (2015), http://

.000000

.000004

.000008

.000012

.000016

.000020

.000024

0

100,000

200,000

300,000

400,000

500,000

600,000

1920 1930 1940 1950 1960 1970 1980 1990 2000 2010

Log (Price Quantity Ratio)US Oil Inventory Quantity

Fig. 2. Logarithm of nominal oil price and US oil inventories: 1920–2014.

5I. Caspi et al. / Energy Economics xxx (2015) xxx–xxx

price increases could again merely be attributed to broader inflationarypressures, according to Hamilton (2011), as real prices remained flat.

The 1970s and early 1980s were, in contrast, periods of great globalstagflation. It also saw great oil shocks emanating from OPEC's oil em-bargo and theYomKippurWar in 1973,which arguably ushered in a de-cade of dramatic oil price increases and supply fears. The Iranianrevolution in 1979, the Iran/Iraq conflict in 1980–1981, and further un-rest in the Middle East during this period, added to the surge in prices.As noted in Hamilton (2011), the shifting of a world petroleummarketcentered in theGulf ofMexico to one centered in the PersianGulf duringthis period, contributed to unprecedented price and supply swings inthe global market for oil.

Our results indicate that the supply ratio shows a protracted periodof price explosiveness during this period, spanning September 1973to April 1986. This significant duration could also be motivated bythe sustained period of global commodity price inflation feedingthrough generally higher oil prices. The real price GSADF suggests twoprotracted periods of explosiveness during this time (for the periods1973M10–1977M03 and 1979M04–1982M0317), followed by a sharpreal price correction in mid 1986 after a significant oil price collapse inlate 1985 (for a more detailed overview of oil prices and supply duringthis period, see Hamilton (2011) and Barsky and Kilian (2002)).

Our estimates suggest no further periods of explosiveness during the1990s for either of our measures following the turbulent two decadespreceding it. A period of stability in oil prices and output ensued,coupled with low global inflation and high growth. The period of lowand stable oil prices (with the real oil price reaching a historic low of$6.92 in December 1998) ended in early 1999, as world petroleum con-sumption increased and the real price of oil rose nearly threefold byFebruary 2000. This was followed by another decline in 2001, followingthe DotCom bubble and subsequent recession in the US. After the startof the second Persian Gulf war in early 2002 and the US deploymentof troops to the region, Middle East tensions drove the oil price uponce again, precipitating a sustained build-up in the real price of oilfor much of the rest of the decade. This was driven, in large part, bysurging global demand for oil, as energy hungry and rampantly growingChina added to the already bulging global oil glut. Fears of Middle Eastproduction shortage continued throughout the decade, as conflicts inthe Niger delta, terrorist attacks in the Persian Gulf, further Arab/Israeli unrest and also the financialization of commodities,18

17 The real and nominal price increases were 91% and 148%, and 53.9% and 108%, respec-tively, for these periods of explosivity.18 Khan (2009), e.g., shows that the daily trading volume of oil futures to world oil de-mand were at 14.7 in 2008, up from approximately 4.5 in 2002. Also see Lombardi andVan Robays (2011) for a discussion on the amplification effect of financialization on theoil price during this time.

Please cite this article as: Caspi, I., et al., Date stamping historical periodx.doi.org/10.1016/j.eneco.2015.03.029

contributed to a massively inflated oil price. Our GSADF results suggestreal oil price explosivity for the period October 2007–August 2008, withreal prices surging 47% (55% nominal) during this time.

The ratio estimates, in contrast, pick up price explosivity earlier, la-beling periods between July and October 2005, April and August 2006,and July 2007 and September 2008 (with the ratio increasing by ap-proximately 16%, 13% and 75% respectively, during these periods). De-spite several supply shocks during this period, global oil productionand US inventories remained relatively stable throughout the 2000s,implying that the latter periods identified could reliably be classifiedas significant price surges relative to supply.

Several sources document the underlying factors to these build-upsof the oil price seen during the periods identified, including Sornetteet al. (2009), Bekiros and Diks (2008), Khan (2009), Lombardi andVan Robays (2011), Areal et al. (2014) and Hamilton (2011). Hamilton(2009) suggests that speculative trading could be regarded as a majorfactor to the increasing price level of commodities in the 2000s, evenwithout any dramatic changes in inventories (as indeed seen duringthe periods identified), given that the short term price elasticity of oildemand is nearly zero. Kilian and Murphy (2014), however, contendsthat speculation among oil traders was not the main driver of theprice surge, stating that the build-up was rather a result of the earliermentioned global surge in demand for oil during this period. This latterview is echoed by most other studies mentioned in this paper, whichidentified similar commodity price bubbles during this period.

In conclusion, although the debate as to the exact causes of theabovementioned periods of explosivity remains open, the impact ofsuch oil price surges are widely accepted. Periods of oil price explosivityare generally regarded as disruptive to production (particularly in in-dustries with high energy needs), consumer demand and price stabilityin oil-importing countries. More research is needed into the macroeco-nomic impact that such periods have on the price stability of moderneconomies, and how and whether policymakers should intervene insmoothing such disruptions. Our estimates can thus be used to serveas a benchmark for analyzing the impact of such periods.

6. Conclusion

The aim of this paper is to provide a basis for effectively labeling his-toric periods of oil price explosivity between 1876 and 2014. We makeuse of Phillips, Shi, and Yu's (2013) right-tailed recursive GSADF ap-proach in order to approximate the starting and ending points of histor-ic oil price explosivity. The GSADF procedure provides an effectivemeans of identifying such periods by recursively adjusting the movingwindow estimation sample for the right-tailed sup ADF testingprocedures. Phillips et al. (2013) show in their paper the ability ofthese techniques to identify multiple periodically collapsing bubbles in

ds of oil price explosivity: 1876–2014, Energy Econ. (2015), http://

6 I. Caspi et al. / Energy Economics xxx (2015) xxx–xxx

a longer dated series. The detection strategy is based on a right-tailedvariant of the standard Augmented Dickey–Fuller (ADF) test, with the al-ternative of a mildly explosive series. The GSADF statistics are comparedto the corresponding calculated critical values, after which we date-stamp these periods using the backwards sup ADF (GSADF) statistic.

Although the technique provides an efficient and consistent basis foridentifying such periods of departures from a unit root process, it pro-vides no causal understanding of such periods. We also apply cautionin not defining in our study such periods as displaying market exuber-ance, or bubble episodes, as is done in most other similar applications.As we lack historical data on derivatives in order to calculate a conve-nience yield, we consider periods of significant oil price build-ups rela-tive to aggregate price levels andUS oil inventory supplies, respectively.

Our results suggest the presence of multiple periods of explosivity inboth the real price and the price supply ratio of oil. In particular, theperiod after the generally stagnant price period between 1941 and1973 displays significantly more explosiveness than before. We findthat during the late 1970s and early 1980s, several periods of prolongedoil price build-ups ensued. We also find, for both measures, that in thebuild-up to the global financial crisis, oil prices experienced explosive-ness (as confirmed by many other studies). The supply ratio indicatesthat prices were significantly above their past stationary levels withrespect to supply even well before then, between 2005 and 2006.

In summary, our study provides a concise estimate of the periodswhere historical oil prices deviated significantly from general price-and US inventory supply levels, since 1876 and 1920, respectively.Although the debate as to the exact causes of the identified periods ofexplosivity remains open, the impact of such oil price surges is widelyaccepted as being disruptive to ordinary economic activity.

Appendix A. Supplementary data

Supplementary data to this article can be found online at http://dx.doi.org/10.1016/j.eneco.2015.03.029.

References

Areal, F.J., Balcombe, K., Rapsomanikis, G., 2014. Testing for bubbles in agriculturalcommodity markets. ESA Working Paperp. 14.

Barsky, R.B., Kilian, L., 2002. Do we really know that oil caused the great stagflation? Amonetary alternative. NBER Macroeconomics Annual 2001. vol. 16. MIT Press,pp. 137–198.

Bekiros, S.D., Diks, C.G., 2008. The relationship between crude oil spot and futures prices:cointegration, linear and nonlinear causality. Energy Econ. 30 (5), 2673–2685.

Please cite this article as: Caspi, I., et al., Date stamping historical periodx.doi.org/10.1016/j.eneco.2015.03.029

Bettendorf, T., Chen, W., 2013. Are there bubbles in the Sterling-dollar exchange rate?New evidence from sequential ADF tests. Econ. Lett. 120 (2), 350–353.

Branch, W.A., Evans, G.W., 2011. Learning about risk and return: a simple model ofbubbles and crashes. Am. Econ. J. Macroecon. 3 (3), 159–191.

Caspi, I., 2013. Rtadf: Testing for Bubbles with EViews.Cuñado, J., Gil-Alana, L.A., Gracia, F.P.d., 2007. Testing for stock market bubbles using non-

linear models and fractional integration. Appl. Financ. Econ. 17 (16), 1313–1321.Diba, B.T., Grossman, H.I., 1988. Explosive rational bubbles in stock prices? Am. Econ. Rev.

78 (3), 520–530.Etienne, X.L., Irwin, S.H., Garcia, P., 2014. Bubbles in food commodity markets: four

decades of evidence. J. Int. Money Financ. 42 (C), 129–155.Evans, G.W., 1991. Pitfalls in testing for explosive bubbles in asset prices. Am. Econ. Rev.

81 (4), 922–930.Gilbert, C., 2010. Speculative influences on commodity futures prices 2006–2008. United

Nations Conference on Trade and Development (UNCTAD) Discussion.Gürkaynak, R.S., 2008. Econometric tests of asset price bubbles: taking stock. J. Econ. Surv.

22 (1), 166–186.Hamilton, J., 2009. Understanding crude oil prices. Energy J. vol. 30 (Number 2), 179–206.Hamilton, J.D., 2011. Historical oil shocks. Technical Report. National Bureau of Economic

Research.Homm, U., Breitung, J., 2012. Testing for speculative bubbles in stock markets: a compar-

ison of alternative methods. J. Financ. Econ. 10 (1), 198–231.Khan, M., 2009. The 2008 oil price “bubble”. Policy Briefs PB09-19. Peterson Institute for

International Economics.Kilian, L., Murphy, D.P., 2014. The role of inventories and speculative trading in the global

market for crude oil. J. Appl. Econ. 29 (3), 454–478.Lammerding, M., Stephan, P., Trede, M., Wilfling, B., 2013. Speculative bubbles in recent

oil price dynamics: evidence from a Bayesian Markov-switching state-spaceapproach. Energy Econ. 36, 491–502.

Lee, J., Strazicich, M.C., 2003. Minimum Lagrange multiplier unit root test with twostructural breaks. Rev. Econ. Stat. 85 (4), 1082–1089.

Lombardi, M.J., Van Robays, I., 2011. Do Financial Investors Destabilize the Oil Price?Lumsdaine, R.L., Papell, D.H., 1997. Multiple trend breaks and the unit-root hypothesis.

Rev. Econ. Stat. 79 (2), 212–218.Phillips, P., Magdalinos, T., 2007. Limit theory for moderate deviations from unit root.

J. Econ. 136, 115–130.Phillips, P.C.B., Yu, J., 2011. Dating the timeline of financial bubbles during the subprime

crisis. Quant. Econ. 2 (3), 455–491.Phillips, P.C.B., Wu, Y., Yu, J., 2011. Explosive behavior in the 1990s Nasdaq: when did

exuberance escalate asset values? Int. Econ. Rev. 52 (1), 201–226.Phillips, P.C., Shi, S.-P., Yu, J., 2013. Testing for multiple bubbles: historical episodes of

exuberance and collapse in the S&P 500. Cowles Foundation Discussion Papers.Pindyck, R.S., 1993. The present value model of rational commodity pricing. Technical

Report. National Bureau of Economic Research.Regnier, E., 2007. Oil and energy price volatility. Energy Econ. 29 (3), 405–427.Reitz, S., Slopek, U., 2009. Non-linear oil price dynamics: a tale of heterogeneous

speculators? Ger. Econ. Rev. 10 (3), 270–283.Shi, S., Arora, V., 2012. An application of models of speculative behaviour to oil prices.

Econ. Lett. 115 (3), 469–472.Sornette, D., Woodard, R., Zhou, W.-X., 2009. The 2006–2008 oil bubble: evidence of

speculation, and prediction. Physica A 388 (8), 1571–1576.

ds of oil price explosivity: 1876–2014, Energy Econ. (2015), http://