Is There Evidence of Super Cycles in Oil Prices?*
Abdel M. Zellou and John T. Cuddington**
March 22, 2012
Abstract
A number of authors have claimed that the strong upward movement in commodity prices since
2000 represents the early phase of a ‘super cycle’ (SC) driven by the sustained rise in demand associated
with industrialization and urbanization in the BRIC countries (Brazil, Russia, India and China).
Cuddington and Jerrett (2008) provide statistical evidence on the presence of super cycles in metals
prices, defining super cycles as cyclical components between 20 and 70 years (trough to trough) using
the Christiano-Fitzgerald band-pass filter.
The purpose of this paper is to address the question: is there evidence of super cycles in crude oil
prices? On one hand, one might expect the strong demand associated with industrialization and
urbanization to affect energy prices in a way that is roughly similar to metals, as both are nonrenewable
resources. On the other hand, the structure of the crude oil market is quite different from that in other
mineral markets, and that structure has changed rather dramatically over time.
Our empirical analysis suggests that there is strong evidence of super cycles in oil prices in the
post-WWII period, and their timing closely matches the SC timing in metals. It appears that the global
economy is currently in an expansionary phase of a super cycle that started from a trough in 1996. For
the pre-WWII period, on the other hand, the evidence for oil price super cycles is weak. Possible
explanations include: (i) oil was economically less important during European and North American
industrialization episodes, (ii) pervasive U.S. regulation, (iii) large supply-side shocks caused by new
discoveries (e.g. in East Texas and later the Middle East), and (iv) periods of oligopolistic price setting
behavior.
* An earlier version of this paper (SPE 147227) was presented at the SPE Annual Technical Conference
and Exhibition held in Denver, Colorado, USA, 30 October-2 November 2011.
** PhD candidate and William J. Coulter Professor of Mineral Economics, Colorado School of Mines,
Golden, CO.
1
I.
Introduction
Since the late 1990s, there has been a strong upward surge in a broad range of commodity prices
(from nonrenewables to grains, softs and livestock), temporarily interrupted by the sharp global
economic downturn in 2008-9. A number of authors have claimed that the strong upward movement in
commodity prices represents the expansionary phase of a ‘super cycle’ (SC) caused by industrialization
and urbanization in the BRIC countries (Brazil, Russia, India and China), especially China. These socalled super cycles have been defined in the literature (Heap 2005; Cuddington and Jerrett 2008) as
having a period between 20 and 50-70 years – including both the expansion and contraction phases of
the cycle. Cuddington and Jerrett (2008) and Jerrett and Cuddington (2008) provide statistical evidence
on the presence of super cycles in metals prices using the asymmetric Christiano-Fitzgerald (ACF) bandpass filter to decompose a number of metal prices into trend and various cyclical components. (See also
the IMF (2010).)
The primary purpose of this paper is to address the question: is there evidence of super cycles in
crude oil prices? We also provide some interesting statistical evidence on very long-term trends.
Regarding super cycles, one might expect the strong demand associated with industrialization and
urbanization to affect energy prices in a way that is roughly similar to metals, as both are nonrenewable
resources. On the other hand, the structure of the crude oil market is quite different from that in other
mineral markets, and that structure has changed rather dramatically over time. Indeed there was
oligopoly at the end of the nineteenth century during the Rockefeller era, U.S. price controls from the
early 1930s through the 1960s, and the OPEC cartel, which was formed in the early 1960s but became
much more assertive in the 1970s. (See Hamilton (2011) and Yergin (1991) for a detailed discussion of
the history of the oil market. A brief chronology is provided in Appendix A.)
Following the methodology that Cuddington and Jerrett (2008) used to study metal prices, this
paper applies the asymmetric Christiano-Fitzgerald (ACF) band-pass filter to real crude oil prices from
1861 to 2010. The objective is to extract a SC cyclical component with a period between 20 and 70
2
years. Cyclical components with periodicity greater than 70 years are then defined as the (very) longterm trend. Short and intermediate-term cyclical components (<20 years) were also extracted, but are
not the focus of this paper. After extracting the SC component in crude oil prices, we ask: is the SC
component in oil prices similar in timing to those for metal prices – as identified in Heap’s (2005)
historical analysis and Cuddington and Jerrett’s (2008) band-pass filter analysis? We also compare our
oil price super cycles to the oil price ‘epochs’ discussed in Dvir and Rogoff (2009) in Appendix B.
The remainder of the paper is organized as follows. Section 2 provides some additional
background and motivation for our analysis. Section 3 discusses the band-pass filter methodology for
decomposing an economic time series into a number of cyclical components with differernt periods (or
conversely frequencies). Section 4 describes our long-span annual data set for crude oil prices covering
the period from 1861 through 2010 and the application of the ACF band-pass filter to obtain various
cyclical components of real crude oil prices, as well as the very long-term trend. These results are
interpreted by referring important facts about the economic history of the oil market and
industrialization episodes in different regions of the world. Section 5 concludes the paper.
II.
Background and Motivation
The economic explanation of the underlying causes of SCs has not yet been formally modeled,
but any complete explanation would clearly focus on the structural transformation that typically
accompanies rising per capita incomes. Simon Kuznets (1973) describes how the initially high share of
agriculture in total employment gradually declines as the manufacturing sector rises with the
industrialization and urbanization that accompanies economic development. As income rises further, the
service sector gains in relative importance, and is ultimately associated with declines in manufacturing
and further declines in agriculture. In the case of the US, for example, the employment share in the
manufacturing sector went from a few percent in 1800 to almost 40% in 1950 before dropping to around
25% at the beginning of this century. During that time period, the employment share in the agricultural
3
sector decreased sharply from 90% to single digits while the service sector increased from about 10% to
70%. These shifts in employment shares were presumably mirrored by similar movements in
expenditure shares (on the demand side). Given that the manufacturing sector is presumably the most
mineral and energy intensive of the three broad sectors (agriculture, manufacturing, and services), one
would expect a prolonged surge in mineral and energy demand during the industrialization and
urbanization process in developing nations.
Whether the sustained strength in mineral demand during industrialization leads to a prolonged
period of higher mineral prices will depend on the supply response, which in turn may be importantly
affected by market structure. On the supply side, the SC hypothesis is predicated on the belief that
short-run metal and energy supply curves are very price inelastic due to capacity constraints (which
could be related to constraints in production, refining or transportation, via pipeline or tankers) and the
long gestation periods for capacity-expanding investment projects. In the long run, however, energy
supply curves are much more price elastic. The key factor when modeling market dynamics is the speed
of adjustment. How long does it take to go from the short run (where the supply curve is near vertical)
to the long run (where it is more or less horizontal, according to most mineral and energy economists)?
2.1 Who cares about the presence or absence of super cycles?
Evidence demonstrating the presence of super cycles in energy commodities is valuable for
national and state governments, financial institutions, and oil and gas companies. At the level of
government, countries that rely on the import or export of energy commodities need to take into account
the presence of super cycles in energy commodities in order to define their revenue and spending
policies. At the firm level, the exploration-development-production-distribution-research-anddevelopment cycle of energy projects often spans several decades, as do super cycles.
Kamel Bennaceur, the chief economist at Schlumberger, one of the world leaders in services in
the oil and gas sector, emphasizes the industry’s long-term planning horizon: “There is high volatility in
the energy sector in terms of prices and investment. Oil and gas companies and service companies work
4
on long-term investment and invest significant amounts of resources on technology and developing
people. Energy investments are for the long term: 30 to 50 years. We need long-term commitments from
both the energy suppliers and the demand sector. We need to project ourselves in 20 years.” This
statement from the World Energy Congress 20101 in Montreal focuses mainly on the supply side and the
timeframe is in the SC range.
III.
Trend-Cycle Decomposition using the Asymmetric Christiano-Fitzgerald Band Pass Filter
Trends and cycles have been widely studied in various subfields in economics. Seasonal
fluctuations, business cycles (6 to 32 quarters), Kitchin inventory cycles (3-5 years), Juglar fixed
investment cycles (7-11 years), Kuznets cycles applied to real estate and infrastructural investment (15
to 25 years), Bronson asset allocation cycles (around 30 years) and Kondratiev waves or “grand super
cycles” (45 to 60 years) are among those that have received attention.
Cuddington and Jerrett (2008) and Jerrett (2010) provide a good description of the asymmetric
Christiano-Fitzgerald band pass filter methodology, which is used in this paper to extract the super
cycles in oil prices. The use of band pass filters in economics has been promoted by Baxter and King
(1999) and Christiano and Fitzgerald (2003).2 The band-pass or frequency filter extracts cyclical
components of a given time series that lie within a specified ‘window’ or range of frequencies or
(conversely) periods. The user specifies the lower and upper bounds of the periods of the cycles of
interest, e.g. cyclical components with periods within the 20-70 year interval. Thinking of frequency
filters in terms of time rather than frequency domain, Baxter and King explain that band-pass filters are
sophisticated two-sided moving averages. They differ from the standard moving averages in two ways.
First the (ideal) weights of various leads and lags are chosen to filter out cyclical components that do not
1
This conference gathers the leading scientists, regulators and business representatives from the energy sector worldwide.
The interview can be found at http://www.energy2point0.com/2010/09/15/kamel-bennaceur-chief-economist-schlumbergerw/
2
Similar band-pass filter techniques are used in different fields, e.g. hard sciences such as electronics and physics. The first
author has encountered it, for example, in spectral imaging and spectral decomposition in geophysics in the oil and gas
industry to extract 3D images of reservoirs in the presence of oil, gas or water.
5
fall within the chosen window. By choosing symmetric weights on each lead and corresponding lag,
phase shift in the extracted component is prevented. Second, there are asymmetric as well as symmetric
filters. Although asymmetric filters invariably introduce some phase shift into the filtered series, they
have the advantage of allowing computation of the filtered series over the entire data span rather than
being limited to a trimmed data span caused by the number of leads and lags used in calculated the
filtered series. This is obviously advantageous if one is particularly interested in studying cyclical
behavior near the end (or beginning) of the available data span.
Table 1 displays our chosen band-pass windows over the entire range of periods [ 2, ∞ ]. Note
that these windows are mutually exclusive and exhaustive, implying that the sum of the components will
sum to the actual series. (The seasonal component is not measurable with annual frequency data.)
Cyclical Component Seasonality Business Cycle Intermediate Cycle Super Cycle Trend SN BC IC SC T Actual Annual Quaterly Monthly Data Data Data NA 2-­‐6 2-­‐18 2-­‐8 6-­‐32 18-­‐96 8-­‐20 32-­‐80 96-­‐240 20-­‐70 80-­‐280 240-­‐840 70-­‐∞ 280-­‐∞ 840-­‐∞ 2-­‐∞ 2-­‐∞ 2-­‐∞ Table 1: Period ‘windows’ for the various cyclical components when using annual, quarterly and
monthly frequency data.
An informal way to assess the plausibility of SCs in crude oil prices is to compare them to the
SCs in metals prices. Since the hypothesis is that the cause of these SCs is urbanization and
industrialization, which are energy and metal intensive, one would expect the SCs in metals and oil
prices to have roughly similar timing that matches known development epochs.3
3
In macroeconomic analysis of cycles, there is no attempt to assess whether any cyclical component is ‘statistically
significant.’ To accomplish this task (for any cyclical component, be it business cycle, seasonal cycle, or super cycle), one
might attempt to devise a method for calculating or simulating confidence bands around cyclical component and then asking
whether this component is ever statistically different from zero. Of course, even a regular sin wave cycle with a period of 70
years, would only have three or four peaks and troughs in a dataset covering 149 years, which would result in a test with very
low power. (See Gurtler and Cumin for a recent suggestion on testing the significance of cyclical component).
6
IV.
A Trend-Cycle Decomposition for Real and Nominal Crude Oil Prices, 1861-2010
4.1
Data and Sources for the Nominal and Real Crude Oil Price
Table 2 summarizes our dataset. The oil price series is from the BP Statistical Review. It spans
over 1861-2010 in annual frequency. To get this long span series, BP splices three different oil price
series.4 From 1861 to 1944, the US average oil price is used. From 1945 to 1985, the oil price from the
Arabian light posted at Ras Tanura is used. Finally from 1986 to 2010, the Brent spot price is used. The
Arabian light series begins in 1945, while the Brent series begins in 1986. An Oregon State University
website publishes longest span U.S. Consumer Price (CPI) Index series starting in 1774 on an annual
basis5.
Table 2: List of Data and Sources
Description
Units
Frequency
Range
Source
BP statistics:
http://www.bp.com/sectiongenericarticle.do?
Oil price
Metals prices
annual
1861-2010
varies with
the metal
PPI
annual
1800-2010
Alan Heap (Citigroup) Database
Carter et al. (2006) then FRED database
(PPIACO)
1774-2010
FRED database and
oregonstate.edu/cla/polisci/downloadconversion-factors
CPI
$/bbl
annual
annual
categoryId=9023773&contentId=7044469
Figure 1 displays the nominal vs. real price of oil using the two different deflators: Producer
Price Index for All Commodities (PPI) and the Consumer Price Index (CPI). The base year used in this
study to compute the real prices is 2005.
4
Appendix C provides unit root tests in the presence of possible breaks at the splice points and also tests for the statistica
significance of breaks. There is no evidence of breaks in the real oil price series at the splice points.
5
A full description of the methodology used to compute the CPI and the different conversion factors are provided in an Excel
spreadsheet, downloadable from oregonstate.edu/cla/polisci/download-conversion-factors
7
Figure 1: Nominal vs. real price of oil using the two different deflators: Producer Price Index for All
Commodities (PPI) and the Consumer Price Index (CPI). The base year is 2005.
4.2
Super Cycles in Nominal and Real Oil Prices
Cuddington and Jerrett (2008) were the first to apply the ACF BPF to commodities, specifically
to metals prices. This paper applies the same statistical technique to oil prices. The ACF filter is used to
extract four mutually exclusive and exhaustive cycles: (i) the business cycle component (2-8 years), (ii)
the intermediate cycle (8-20 years), (iii) the super cycle component (20-70 years) and (iv) the long-term
trend (defined here as cyclical components with periods in excess of 70 years). Our analysis focuses on
the two latter components. If we sum the first three components, we obtain the component BP(2 − 70) .
Substracting this sum from the actual series leaves the very long-term trend BP (70 − ∞) as a residual.
This can be expressed with the following equation:
Actual ≡ [ BC (2,8) + IC (8, 20) + SC (20, 70)] + T (70, ∞)
Actual ≡
BP(2, 70)
+ T (70, ∞)
The calculated cyclical components are shown in Appendix D for interested readers [Attached.
Not for publication. Available on request from the authors.] A graphical presentation of the
components is more revealing. The bottom panel in Figure 2 displays the super cycle component with
the peaks (P) and the troughs (P). The notation P? and T?, on the other hand, denotes the points that are
8
more uncertain due to their low amplitude. It is unclear whether these should be categorized a SC
turning points.
Figure 2: Trend and cycle decomposition of the real price of oil. The top panel displays the real price of
oil, the trend component and the cyclical component using the band pass filter over the 2-70 year period
(BP(2,70)). That cyclical component, BP(2,70), is the summation of the business, intermediate and
supercycle components displaied in the three bottom panels. The bottom panel displays the super cycle
component with the peaks (P) and the troughs (P). P? and T? denote the points that one may or may not
want to classify as SC truing points due to their low amplitude.
One might wonder if the SCs displayed in Figure 2Error! Reference source not found. depend on
the price deflators used, i.e. Producer Price Index for All Commodities (PPI) vs. the Consumer Price
9
Index (CPI). Figure 3 displays the SCs for real price of oil calculated using the alternative deflators.
Note that the SCs for the nominal and real price of oil are very similar (the correlation coefficient is 0.96
when using the CPI and 0.95 when using the PPI as a deflator). Thus, we can conclude that the SCs in
real oil price are not due to movements in the deflator. Moreover, using the PPI or CPI as a deflator
gives very similar results. The difference occurs mainly during the price control period in the 1930s.
Table 3 shows the correlation coefficient matrix between the SCs of the real price of oil, the
nominal price of oil, and the PPI. We can note that the SC for the nominal and real price of oil have a
very high correlation of 0.95, while the correlation coefficient of the nominal price of oil and the price
deflator is 0.79. Appendix C [attached. Not for publication. Available on request from the author]
provides an analysis of the comparison of the WTI and Brent nominal prices, the unit root tests (which
inform the choice of parameters in the ACF filter’s detrending method), and structural break tests at the
splicing dates of the BP price series.
There is strong statistical evidence of three SCs in oil prices: SC1 from 1861 through 1884, SC2
from 1966 through 1996, and SC3 from 1996-present. These are highlighted in Figure 4 below. The
period 1884-1966 is harder to interpret and has been aggregated into one uncertain period for three main
reasons: (i) oil was economically less important during European and North American industrialization
episodes (Figure 5 from Tol (2006)), (ii) the 1884-1966 period includes the price control era, (iii) there
is less of a match in the SCs depending on the deflator used, (iv) there is a lower variance in the price of
oil during that period. The shading corresponds to the super cycles in real oil prices with the
corresponding dates (from trough to trough). The correlation between the SCs in oil and metals prices
before WWII is 0.71 and 0.88 after WWII. It shows a strong correlation between the two after WWII.
10
Figure 3: Super cycles in nominal and real price of oil using CPI and PPI (base year 2005) as price
deflators. There is little difference between the two series, except during the price control period in the
1930s. The shading corresponds to the expansionary phase of the super cycles in real oil prices using the
CPI as deflator. The shading corresponds to the different super cycles in real oil prices as defined in
Figure 2.
Figure 4: Super cycles for the real oil price and metals prices. The units on the vertical axis represent
percentage deviations from trend. For example, +0.40 indicates 40% above the long-term trend (shown
in Figure 2 below). The shading corresponds to the super cycles in real oil prices with the
corresponding dates (from trough to trough). There appear to be three obvious SCs in oil prices, the first
one between 1861 and 1884, and the last two between 1966-1996 and 1996 to date. The period 18841966 is harder to interpret and has been aggregated into one uncertain period.
11
Figure 5: Primary energy consumption level (top panel) and share (bottom panel) for the United States
between 1850 and 2010. These graphs show that the consumption share for oil was low during the
uncertain period. Source: Tol (2006).
12
We investigated the correlation between the SC in the nominal and real price of oil and found a
correlation coefficient of 0.95 before World War II and 0.97 for the period after World War II (Table 3).
Correlation Real Nominal Deflator Real 1 0.95 0.54 Nominal 0.97 1 0.77 Deflator 0.26 0.48 1 Table 3: Correlation Coefficient for super cycles in nominal, real oil prices and the deflator (CPI).6 The
terms in italics above the diagonal represent the correlation coefficients after WWII. The lower left
corner represent the correlation coefficients before WWII.
Figure 6: Real oil prices, along with their trend and super-cycle components. The shading corresponds to
the different super cycles in real oil prices identified in Figure 4 above. An upward trend in real oil
prices started during World War II. In real terms, the trend in oil prices has increased by roughly 125%
over the past 65 years, representing an average annual increase of about 2%. Note that the SC
component was only 20% above the trend in 2010, whereas the earlier SC peak in 1981 was almost
100% above the long-term trend. This suggests the current SC is far from over.
4.3
Comparing the Super Cycles in Oil and Metal Prices
The dating of the super cycles in oil prices can now be compared to the ones in metals. Figure 6
6
All the correlation coefficients in this table and in the following correlation coefficient table are statistically significant at
the 5% level unless marked otherwise.
13
displays the supercycle component and the trend in real oil prices. In this figure the super cycle in oil
price is overlaid with the the principal component of the super cycle for the six LME metals in
(Cuddington and Jerrett 2008). There is a high correlation (.73) between the oil and metals super cycles,
which supposts the SC hypothesis.
The expansionary phase of the SCs can be related to historical episodes of industrialization.
Cuddington and Jerrett (2008) suggest that the SCs in metals prices that they identify are associated with
the industrialization in the US in the late 19th, early 20th century, the reconstruction of Europe after
World War II, the Japanese renaissance in the 1960s and finally the industrialization in the BRIC
countries, mainly driven by China in the 1990s. These SCs in metals match the ones of oil as you can
see in Figure 7. The expansionary phase of the last super cycle started in 1996 with a trough about 70%
below the trend line and the SC component being about 20% above the trend line in 2010. This
information leads us to conclude that we are currently in an expansionary phase of a SC.
Trough year Expansion Phase ~1850 (?) 1884 ~1850-­‐1869 1884-­‐1896 1906 1932 1906-­‐1919 1932-­‐1944 1966 1966-­‐1981 SC SC 1: ~1850-­‐
1884 Uncertain Period : 1884-­‐1966 SC 2: 1966-­‐
1996 1996 1996-­‐? SC 3: 1996-­‐? SC for metals ~1900-­‐1937 1937-­‐1965 1965-­‐1999 1999-­‐? Table 4: Period in oil price super cycles over the period 1861-2010. This table displays the expansionary
phase (from trough to peak) of the SCs in oil prices, as well as the entire SCs (from trough to trough).
For comparison purposes, the SCs for metals are given.
14
Figure 7: Super cycle and trend in real oil prices using the CPI (base year 2005) as price deflator. In the
lower panel, the super cycle in oil price is overlaid with the super cycle in metal prices, defined by the
principal component of the SCs in the six LME metals in Cuddington and Jerrett (2008). There is a
correlation of 0.73 between the oil and the metals super cycles. The right axis is for the trend and the
real price of oil, while the left axis is for the SCs. The shading corresponds to the different super cycles
in real oil prices as defined in Figure 2.
4.4
The Very Long-Term Trend in Real Oil Prices
Figure 5Error! Reference source not found. shows a pronounced upward trend in real oil prices
since World War II. In real terms, the trend component in oil prices more than doubled with an increase
of 126% over the past 65 years, representing an average annual increase of about 2%. Moreover the SC
component was about 20% above the trend in 2010, knowing that the highest upswing was witnessed in
1981 with a magnitude of almost 100%. The trend component presumably reflects the opposing effects
of increasing scarcity due to depletion and ongoing technological change to alleviate scarcity. An
increase in scarcity in oil causes an increase in the price of oil in real term as described by Hotelling
(1931). An improvement in technology reduces cost and increases the reserves that can be developed
from a given resource. Between 1860 and World War II, we witness a downward trend, which may be
interpreted as oil being more abundant with the discovery of major oil fields and technological change
15
with the first logging tool being used in the 1920s for instance. On the other hand, the upward trend
after World War II suggests that increasing scarcity was not completely offset by technological advance.
V.
Conclusions and Extensions
Our band-pass filter analysis and reflections on how to interpret the results suggest the following
conclusions:
•
There is strong evidence of super cycles in oil prices in the post-WWII period, and their
timing closely matches the SC timing in metals. We have identified two earlier super cycles in
oil price: 1861-1884 (trough to trough) and 1966-96. Moreover, it appears that the global
economy is currently in an expansionary phase of the super cycle from the trough around 1996.
This expansionary phase is presumably driven by urbanization and industrialization in the BRIC
countries.
•
For the pre-WWII period, on the other hand, the evidence for oil price super cycles is
weak. Possible explanations include: (i) oil was economically less important during European
and North American industrialization episodes, (ii) pervasive U.S. regulation, (iii) large supplyside shocks caused by new discoveries (e.g. in East Texas and later the Middle East), and (iv)
periods of oligopolistic price setting behavior.
•
The timing of the last two crude oil price super cycles roughly corresponds to the timing
of SCs in metals prices, which in turn matches the timing of the industrialization and
urbanization phase of economic development in US, then Europe and finally Asia.
•
Beyond the SC frequency, there is an upward long-term trend in the real price of oil that
started during World War II. It has averaged 2% per year over the past 65 years, but this trend
has been obscured by super cycles as well as shorter-term business cycle fluctuations.
16
Presumably, these very long-term trends reflect the opposing effects of increasing scarcity (due
to depletion) and ongoing technological change (to alleviate scarcity through more efficient use
of existing reserves and/or new discoveries).
Our ongoing research on super cycles has empirical and theoretical aspects. On the empirical
front, we are in the process of extending our analysis to coal and natural gas. Preliminary analysis of
coal suggests that there is a strong correlation between the SCs in oil and coal prices, in spite of very
different market structures for the two energy products. On the theoretical front, we are attempting to
develop a simple theoretical model capable of generating super cycle behavior. (See Cuddington and
Zellou (2012)).
In the meantime, we hope our empirical investigation of the possibility of super cycles in crude
oil prices will be of interest to energy-sector analysts, the investment community, and governments in
both energy exporting and importing countries coping with changing trends and cycles in their terms of
trade, and their implications for government revenues.
References
Baxter, Marianne, and Robert G. King. "Measuring Business Cycles: Approximate Band-Pass Filters for
Economic Time Series." Review of Economics and Statistics 81, no. 4 (1999): 575-93.
BP, British Petroleum. "BP Statistical Review." http://www.bp.com/statisticalreview.
Christiano, Lawrence, and Terry Fitzgerald. "The Band Pass Filter." International Economic Review 44,
no. 2 (2003): 435-65.
Comin, Diego, and Mark Gertler. "Medium-Term Business Cycles." American Economic Review 96, no.
3 (2006): 523-51.
Cuddington, John T., and Daniel Jerrett. "Super Cycles in Real Metals Prices?". IMF Staff Papers 55,
no. 4 (2008): 541-65.
Cuddington, John T., Rodney Ludema, and Shamila A. Jayasuriya. "Prebisch-Singer Redux." In Natural
Resources: Neither Curse Not Destiny, edited by Daniel Lederman and William F. Maloney.
Stanford University Press, 2007.
Cuddington, John T., and Carlos M. Urzua. "Trends and Cycles in the Net Barter Terms of Trade: A
New Approach." Economic Journal 99, no. 396 (1989): 426-42.
17
Cuddington, John T., and Abdel M. Zellou. "A Simple Mineral Market Model: Under What Conditions
Will It Produce Super Cycles in Prices?" Colorado School of Mines, 2012.
Deaton, Angus, and Guy Laroque. "Competitive Storage and Commodity Price Dynamics." The Journal
of Political Economy 104, no. 5 (1996): 896-923.
———. "On the Behaviour of Commodity Prices." The Review of Economic Studies 59, no. 1 (1992): 123.
Dvir, Eyal, and Kenneth S. Rogoff. "Three Epochs of Oil." National Bureau of Economic Research
Working Paper Series No. 14927 (2009).
Hamilton, James D. "Historical Oil Shocks." National Bureau of Economic Research Working Paper
Series No. 16790 (2011).
Heap, Alan. "China - the Engine of a Commodities Super Cycle." edited by Equities Research: Global.
New York City: Citigroup Smith Barney, 2005.
IMF, International Monetary Fund. "World Economic Outlook October 2010." 51, 2010.
Jerrett, Daniel. "Trends and Cycles in Metals Prices." PhD dissertation (Mineral and Energy
Economics), Division of Economics and Business. Colorado School of Mines, 2010.
Jerrett, Daniel, and John T. Cuddington. "Broadening the Statistical Search for Metal Price Super Cycles
to Steel and Related Metals." Resources Policy 33, no. 4 (2008): 188-95.
Kuznets, Simon. "Modern Economic Growth: Findings and Reflections." American Economic Review
63, no. 3 (1973): 247-58.
Tol, Richard S., Stephen W. Pacala , and Robert Socolow. "Understanding Long-Term Energy Use and
Carbon Dioxide Emissions in the Usa." SSRN eLibrary (2006).
Yergin, Daniel. The Prize: The Epic Quest for Oil, Money, and Power. New York: FreePress, 1991.
18
APPENDIX A: A Brief History of Oil
Hamilton (2011) and Yergin (1991) provide a very detailed description of the history of the price of oil.
The brief chronology provided here relies mainly on these two sources.
-­‐
1859-1899: Let there be light
o 1862-1864: the first oil shock with the rapid drop in oil prices
o 1865-1899: evolution of the industry: still drop in price
 Hamilton does not see agree with the interpretation of Dvir and Rogoff on the
similarities in the behavior of oil prices, in terms of restriction of access to the
excess oil supply, between the 19th century and the last quarter of the 20th century.
Indeed, Hamilton argues that oil did not have as much economic importance at the
end of the 19th century compared to the end of the 20th century. The share of oil in
GNP is much smaller in the 19th century compared to the last quarter of the 20th
(0.4% of 1900 GNP compared to 4.8% of 2008 GDP).
-­‐
1900-1945: Power and transportation
o The west coast gasoline famine of 1920
o The great depression and state regulations. These state regulations focused on restricting
production, which allowed a better management of the East Texas Oil Field compared to
the early fields in Pennsylvania (see Figures 1 and 5 of the cited paper).
-­‐
1946-1972: The early postwar era.
o State regulation
o 1947-1948: Postwar dislocations: increase in the price of oil due to acceleration in the use
of vehicles.
o 1952-1953: supply disruptions and the Korean conflict.
o 1956-1957: Suez Crisis
o 1969-1970: Modest price increases.
-­‐
1973-1996: The age of OPEC.
o 1973-1974: OPEC Embargo.
o 1978-1979: Iranian revolution.
o 1980-1981: Iran-Iraq War.
o 1981-1986: The great price collapse.
o 1990-1991: First Persian Gulf War.
-­‐
1997-2010: A new industrial age.
o 1997-1998: East Asian Crisis.
o 1999-2000: Resumed growth.
o 2003: Venezuelan unrest and the second Persian Gulf War.
o 2007-2008: Growing demand and stagnant supply.
19
APPENDIX B: Comparison to Dvir and Rogoff
Dvir and Rogoff (DR) perform a statistical analysis of the change in the persistence and the
change in volatility of the price of oil, provide an extended commodity storage model, and give a
historical explanation of these transition points related to industrialization and a change in market
structure.
DR analyze the same long-span real oil price series considered in this paper. At the outset, they
note that “Studies of the time series properties of real oil prices have taken one of the following
approaches in the face of these clear non-linearities in the series: either analyzing the series as a whole,
or, much more commonly, treating the series as composed of separate series ‘pasted together’, and
proceeding to analyze them in isolation.” They pursue the second approach, using a number of
econometric techniques to test for possible structural breaks at unknown dates. Their tests allow for
changes in persistence and volatility of the oil price series: “In what follows we will treat both the
assumption of a pure I(0) process and the assumption of a pure I(1) process as our null hypotheses, and
test whether the series exhibits a shift from I(0) to I(1) (or vice versa) against both of these assumptions.
[…] This allows us to test for structural change without taking an a-priori stand regarding the null
hypothesis.”
In contrast, our approach of looking for very-long cycles requires that we consider the price
series as a whole. One cannot detect 20-70 year cycles, in subsample of the data span that are of
relatively short duration. The BPF analysis requires the specification of the nature of the underlying
trend: mean stationarity, trend stationarity or a unit root process with drift. Based on our unit root test
results, we adopted the latter specification. Clearly, this is less general than the DR approach of
allowing for shifts in the persistence of the oil price series from I(0) to I(1), but unavoidable if one wants
to employ existing BPF methodology.
Dvir and Rogoff detect three ‘epochs’ in oil prices characterized by changes in persistence and
volatility:
•
•
•
Epoch I from 1861-1877 has high prices and high volatility
Epoch II from 1878-1972 is a period of low prices with low (1878-1933) and then even lower
(1934-72) volatility
Epoch III from 1973-2009 is again a period of high prices and high volatility.
Figure 7 displays the three epochs from DR with the shading overlaid on our oil price SCs. For
consistency, we use the CPI as the deflator, as in DR. There is a similarity between the epochs defined
by DR and the SCs. The dating of their epochs matches up well with the neutral zero value for the super
cycle component. We define our super cycle expansions going from trough to peak, whereas the DR
epochs in effect go from our SC midpoint to midpoint. Interestingly, the BPF analysis identifies the
start of a new SC emerging from a trough around 1996.
DR emphasize that their epochs reflect a confluence of strong demand growth (associated with
industrialization in major regions of the world) and important changes in market structure on the supply
side. They stress the pivotal role of supply restrictions in leading to the high volatility and high prices in
Epoch I and Epoch III when industrialization was occurring in the US and South East Asia, respectively.
During Epoch I, the restriction in supply was initiated by Rockefeller with the quasi monopoly on oil
refining and the oligopoly of the railroad companies in the transportation of oil. In Epoch III, the
restriction of supply was due to the rising market power of OPEC.
20
The amplitude of our first and last two SCs is much more pronounced than in the middle where
the SCs are affected by price regulation. In that respect, we do see similarities if we take into account
the amplitude of these SC and superimpose it on the three epochs defined by DR. DR used a different
statistical technique and we arrive at similar conclusion in terms of the price behavior of oil in the longterm.
Figure 8: Super cycles on real prices of oil (2005) (CPI deflated) with the epochs defined by Dvir and
Rogoff (shaded in this graph).
21
***** Not for publication. Available on request from the authors.
APPENDIX C: Econometric Analysis of the BP Price Series
Comparison of the WTI and Brent Nominal Prices
Figure 8 displays the nominal price of oil (on natural logarithm scale). In order to assess if the
choice of benchmark (Brent, WTI, Arabian light, etc.) is impacting the analysis, we decided to make a
comparison between WTI and Brent over the 1986-2010 period. Figure 9 gives the price of oil in log
scale using WTI. Figure 10 shows both prices (Brent and WTI) in logs as well as the difference. There
is no significant difference between WTI and the Brent prices. Hence, the results presented in this paper
will be using the BP series only, even though we computed the SCs in both cases using Brent and WTI.
Figure 9: Nominal price of oil in level on a log scale (Source BP statistical review (BP 2011)).
Figure 10: Nominal price of oil in level on a log scale using the WTI starting in 1986.
22
Figure 11: Nominal price of the WTI and Brent in logs over the 1986-2010 period and the difference.
Unit Root Tests
As described earlier, the oil series is composed of three different prices, with the Arabian Light
price being used after 1944 and the Brent price starting in 1986 and. We want to investigate whether
there is a presence of a structural break at these splicing points (at 1986 and 1944).
First, we checked the presence of a unit root on the log of the real price of oil. The PhillipsPerron test statistic in Table 6 shows that the log of the real price of oil is integrated of degree one
(I(1)). It means that the time series needs to be differentiated once to be stationary. Failing to do so will
lead to spurious regressions as described by Cuddington, Ludema, and Jayasuriya (2007) and
Cuddington and Urzua (1989). The Philips-Perron unit root test is preferred to the Dickey-Fuller test in
the case of time varying volatility, which is the case here. The Philips-Perron unit root test uses robust
standard error while the Dickey-Fuller does not. The null hypothesis of a second unit root is rejected.
The lag length selection criteria show that two lags are necessary based on the sequential modified
likelihood ratio (LR) test statistic and the Akaike information criterion (AIC) (Table 7). Hence, we will
run the following univariate model:
DLPRoil _ BP _ cpit =α + β1 DLPRoil _ BP _ cpit −1 + DLPRoil _ BP _ cpit −2 + et
23
Table 5: Unit root test on the real price of oil. The series is I(1)
Table 6: Lag length selection for the real price of oil
Structural Breaks
When using the univariate equation specified above to test for the presence of structural breaks at
the splicing date we find that there are no structural breaks (Table 8, Figure 12 and Table 9). We also
performed a Chow breakpoint test at several other dates around these splicing dates, in 1945, 1946, 1984
and 1985, and they all rejected the presence of a structural break.
24
Table 7: Univariate regression of the real price of oil using two lags
Table 8: Chow test to test for structural breaks at the splicing points of the oil price series in 1944 and
1986. There are no structural breaks
Figure 12: Correlogram of the residual corresponding to the univariate regression of the real oil price
above (in first difference of the log term)
25
***** Not for publication. Available on request from the authors.
APPENDIX D: Trend and Cycles components for Real Oil Prices Based on the Asymmetric CristianoFitzgerald Band-Pass Filter
All these series are in log terms.
Real Oil Price obs 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 NA 3.20 3.93 4.93 5.73 5.28 4.64 4.19 4.61 4.65 4.74 4.90 4.74 4.04 3.62 3.76 4.42 4.42 3.73 3.40 3.51 3.40 3.26 3.48 3.36 3.45 3.26 3.19 3.47 3.51 3.43 3.17 3.01 3.13 Business Cycle Trend NA 3.73 3.75 3.77 3.79 3.80 3.82 3.83 3.84 3.85 3.86 3.87 3.87 3.88 3.88 3.88 3.88 3.88 3.87 3.87 3.86 3.85 3.84 3.83 3.82 3.80 3.79 3.77 3.75 3.73 3.71 3.69 3.66 3.64 Intermediate Cycle NA -­‐0.58 -­‐0.31 0.29 0.82 0.27 -­‐0.35 -­‐0.68 -­‐0.12 0.05 0.22 0.44 0.34 -­‐0.28 -­‐0.62 -­‐0.36 0.42 0.54 -­‐0.04 -­‐0.25 -­‐0.03 -­‐0.05 -­‐0.11 0.15 0.04 0.11 -­‐0.11 -­‐0.20 0.08 0.16 0.14 -­‐0.08 -­‐0.23 -­‐0.16 Super Cycle NA -­‐0.13 0.17 0.41 0.53 0.52 0.38 0.18 -­‐0.01 -­‐0.15 -­‐0.22 -­‐0.24 -­‐0.22 -­‐0.18 -­‐0.15 -­‐0.11 -­‐0.07 -­‐0.03 0.01 0.04 0.07 0.08 0.09 0.11 0.13 0.16 0.18 0.16 0.10 -­‐0.01 -­‐0.13 -­‐0.24 -­‐0.30 -­‐0.30 NA 0.19 0.32 0.46 0.58 0.70 0.79 0.85 0.89 0.90 0.88 0.82 0.74 0.63 0.50 0.35 0.19 0.04 -­‐0.12 -­‐0.26 -­‐0.38 -­‐0.49 -­‐0.56 -­‐0.61 -­‐0.63 -­‐0.62 -­‐0.59 -­‐0.53 -­‐0.46 -­‐0.38 -­‐0.29 -­‐0.20 -­‐0.12 -­‐0.05 26
1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 3.43 3.94 3.78 3.37 3.51 3.86 3.75 3.50 3.29 3.42 3.29 2.97 3.13 3.09 3.03 3.04 2.90 2.85 3.04 3.23 3.09 2.84 3.37 3.64 3.69 3.50 3.77 3.03 3.05 2.93 2.98 3.14 3.23 2.85 2.76 2.86 2.79 2.21 2.59 2.42 2.87 2.81 2.90 2.97 3.62 3.59 3.56 3.54 3.51 3.48 3.46 3.43 3.40 3.37 3.34 3.31 3.29 3.26 3.23 3.20 3.18 3.15 3.13 3.10 3.08 3.05 3.03 3.01 2.99 2.97 2.95 2.93 2.91 2.89 2.88 2.86 2.85 2.84 2.82 2.81 2.80 2.79 2.79 2.78 2.77 2.77 2.76 2.76 0.04 0.43 0.15 -­‐0.35 -­‐0.23 0.14 0.12 -­‐0.02 -­‐0.11 0.14 0.09 -­‐0.17 0.03 0.01 -­‐0.02 0.03 -­‐0.06 -­‐0.08 0.12 0.25 0.00 -­‐0.40 -­‐0.03 0.11 0.10 -­‐0.04 0.34 -­‐0.25 -­‐0.09 -­‐0.11 -­‐0.03 0.14 0.22 -­‐0.13 -­‐0.13 0.09 0.16 -­‐0.32 0.09 -­‐0.12 0.22 0.03 0.01 0.00 -­‐0.24 -­‐0.13 -­‐0.01 0.10 0.17 0.20 0.19 0.15 0.10 0.05 0.03 0.02 0.03 0.02 0.00 -­‐0.05 -­‐0.12 -­‐0.19 -­‐0.23 -­‐0.23 -­‐0.16 -­‐0.04 0.10 0.21 0.26 0.23 0.14 0.03 -­‐0.06 -­‐0.10 -­‐0.07 -­‐0.01 0.06 0.09 0.06 -­‐0.02 -­‐0.13 -­‐0.21 -­‐0.23 -­‐0.18 -­‐0.08 0.05 0.15 0.20 0.01 0.05 0.07 0.07 0.06 0.03 -­‐0.01 -­‐0.05 -­‐0.10 -­‐0.14 -­‐0.17 -­‐0.20 -­‐0.21 -­‐0.20 -­‐0.18 -­‐0.14 -­‐0.09 -­‐0.03 0.03 0.10 0.17 0.23 0.28 0.32 0.34 0.35 0.34 0.32 0.29 0.24 0.20 0.14 0.09 0.05 0.01 -­‐0.02 -­‐0.05 -­‐0.06 -­‐0.06 -­‐0.06 -­‐0.05 -­‐0.03 -­‐0.02 0.00 27
1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 2.89 2.81 2.82 2.92 2.92 2.81 2.75 2.59 2.64 3.08 2.97 2.77 2.75 2.73 2.65 2.75 2.74 2.74 2.74 2.71 2.77 2.74 2.64 2.57 2.56 2.55 2.54 2.52 2.51 2.48 2.45 2.40 2.35 2.51 2.56 2.81 4.01 3.89 3.89 3.92 3.86 4.59 4.63 4.46 2.76 2.76 2.76 2.76 2.76 2.76 2.76 2.76 2.77 2.77 2.78 2.78 2.79 2.80 2.81 2.81 2.82 2.83 2.84 2.86 2.87 2.88 2.89 2.91 2.92 2.94 2.95 2.97 2.98 3.00 3.02 3.03 3.05 3.07 3.09 3.11 3.13 3.15 3.17 3.20 3.22 3.24 3.27 3.29 -­‐0.09 -­‐0.13 -­‐0.05 0.12 0.16 0.06 -­‐0.01 -­‐0.21 -­‐0.20 0.23 0.12 -­‐0.04 -­‐0.03 0.00 -­‐0.06 0.05 0.02 -­‐0.01 -­‐0.02 -­‐0.05 0.03 0.06 0.01 -­‐0.02 -­‐0.01 -­‐0.02 -­‐0.03 -­‐0.03 0.00 0.04 0.08 0.07 -­‐0.04 -­‐0.03 -­‐0.22 -­‐0.28 0.59 0.17 -­‐0.06 -­‐0.22 -­‐0.41 0.23 0.22 0.05 0.20 0.14 0.06 -­‐0.01 -­‐0.06 -­‐0.07 -­‐0.05 -­‐0.02 0.02 0.03 0.03 0.00 -­‐0.04 -­‐0.08 -­‐0.11 -­‐0.10 -­‐0.08 -­‐0.03 0.01 0.05 0.06 0.06 0.05 0.06 0.09 0.13 0.16 0.15 0.10 -­‐0.01 -­‐0.14 -­‐0.26 -­‐0.33 -­‐0.31 -­‐0.23 -­‐0.09 0.05 0.16 0.22 0.24 0.23 0.22 0.19 0.15 0.02 0.04 0.05 0.06 0.06 0.06 0.06 0.06 0.05 0.05 0.04 0.03 0.03 0.02 0.01 -­‐0.01 -­‐0.03 -­‐0.06 -­‐0.09 -­‐0.14 -­‐0.19 -­‐0.25 -­‐0.32 -­‐0.38 -­‐0.44 -­‐0.50 -­‐0.54 -­‐0.56 -­‐0.57 -­‐0.55 -­‐0.51 -­‐0.44 -­‐0.34 -­‐0.22 -­‐0.08 0.08 0.24 0.40 0.56 0.70 0.81 0.90 0.95 0.97 28
1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 4.26 4.09 4.03 3.94 3.26 3.48 3.24 3.39 3.61 3.38 3.30 3.14 3.04 3.09 3.26 3.15 2.72 3.05 3.49 3.30 3.29 3.42 3.68 4.00 4.15 4.22 4.51 4.03 4.26 3.31 3.34 3.36 3.39 3.41 3.44 3.47 3.49 3.52 3.54 3.57 3.60 3.63 3.65 3.68 3.71 3.73 3.76 3.79 3.82 3.84 3.87 3.89 3.92 3.95 3.97 4.00 4.02 4.04 -­‐0.08 -­‐0.10 0.05 0.19 -­‐0.29 0.08 -­‐0.11 0.05 0.23 -­‐0.01 -­‐0.05 -­‐0.13 -­‐0.12 0.02 0.25 0.14 -­‐0.33 -­‐0.08 0.28 0.01 -­‐0.10 -­‐0.11 -­‐0.02 0.11 0.07 -­‐0.02 0.19 -­‐0.30 -­‐0.04 0.08 -­‐0.03 -­‐0.16 -­‐0.29 -­‐0.37 -­‐0.36 -­‐0.27 -­‐0.11 0.06 0.21 0.28 0.28 0.22 0.13 0.06 0.00 -­‐0.03 -­‐0.06 -­‐0.09 -­‐0.13 -­‐0.16 -­‐0.16 -­‐0.11 -­‐0.03 0.06 0.13 0.15 0.11 0.04 0.94 0.88 0.78 0.65 0.50 0.33 0.15 -­‐0.03 -­‐0.21 -­‐0.36 -­‐0.50 -­‐0.60 -­‐0.68 -­‐0.72 -­‐0.73 -­‐0.70 -­‐0.65 -­‐0.58 -­‐0.49 -­‐0.39 -­‐0.29 -­‐0.18 -­‐0.09 0.00 0.07 0.13 0.17 0.20 0.21