Will the energy price bubble burst?

Energy 150 (2018) 276e288

Contents lists available at ScienceDirect

Energy journal homepage: www.elsevier.com/locate/energy

Will the energy price bubble burst? Tie-Ying Liu a, Chien-Chiang Lee b, * a b

School of Economics and Management, Beijing Jiaotong University, Beijing, China Department of Finance, National Sun Yat-sen University, Kaohsiung, Taiwan

a r t i c l e i n f o

a b s t r a c t

Article history: Available online 19 February 2018

In this paper, we employ recursive unit-root tests to identify the beginnings and ends of potential speculative bubbles in the energy price market. The major advantage of this approach is that it allows one to account for a nonlinear structure and break while investigating the possible existence of multiple structural breaks resulting from these bubbles. The evidence finds the efficient market hypothesis has been the failure in most of the energy market. It shows that the gasoline spot market price is volatile, and investors in gasoline markets have more speculative opportunities than those investors who enter coal and oil markets. It is possible to hedge benefits in oil spot and future markets. Furthermore, the coalgasoline price relation as well as the coal-crude oil price relation has a weak link, and the existing bubbles influence the interactions between the oil and gasoline markets. © 2018 Elsevier Ltd. All rights reserved.

JEL: C22 Q43 Q48 Keywords: Energy price Multiple bubbles Nonlinear structure Sup ADF test Generalized sup ADF test Speculative opportunities

The fluctuation of the energy price affects industry development and plays an important role in terms of national investment structures, economic performance, and inflation rates [1]. The energy price has experienced a dramatic rise, followed by a spectacular crash, and this fluctuation may indicate that the energy price can systematically deviate from its fundamental value. Hence, it is essential to analyze the causes of energy price bubbles. Investigating whether speculative behavior in futures markets contributes to price bubbles in physical markets is an important issue for the current academic debate, as it is of public concern. The importance of energy resources to an economy implies that rigorous and reliable forecasts the energy price should be much invaluable to policymakers. It is plausible that key macroeconomic variables may inherit the stochastic properties of energy commodities. Therefore, understanding the true nature of energy prices variability should be a critical issue to the policy-making process [2]. Indeed, energy prices influence the level of a nation capital expenditure and promote a nation incentive to find alternative energy sources. Investors in energy markets may find that energy

prices tend to reach a relatively certain level or parity as the basis for a trading strategy [3].1 Many people may want to see if we can use the explosive patterns in energy prices as a leading indicator for future output or financial market conditions. Or these big changes in prices just reflect the big changes in supply. Crude oil prices are notoriously volatile in the past decades. Crude oil prices have experienced a historic pattern of what we call directional volatility, where prices are volatile in one direction (up or down) for a four months prolonged period. The standard example of directional crude-price volatility occurred from the mid-1970s to the mid1980s, and it consisted of the first and second oil price shocks and the subsequent price declines [4]. In the past few years, the international crude oil price has experienced extreme volatility due to the confluence of numerous risk factors, which has led to large uncertainties in oil price forecasts and market risks for investors [5]. Since 2004, international crude oil prices have started to increase in the energy market and reached a record high of $147 USD per barrel until 2008 [6]. Then, oil prices exhibited a spectacular crash from the historical high on July 11, 2008 to a low of $33.87 USD on December 19, 2008 [7]. The global diesel and gasoline demand for road transportation has consistently grown faster than

* Corresponding author. Department of Finance at National Sun Yat-sen University, Taiwan. E-mail address: [email protected] (C.-C. Lee).

1 The Atil et al. [3] study offers both investors and the government economic implication insights through the fluctuations of these energy-based commodity prices.

1. Introduction

https://doi.org/10.1016/j.energy.2018.02.075 0360-5442/© 2018 Elsevier Ltd. All rights reserved.

T.-Y. Liu, C.-C. Lee / Energy 150 (2018) 276e288

the total oil product consumption in the past four decades, rising from approximately 30% of oil product consumption in 1971 to approximately 46% in 2007 [8]. In the past two decades, we have witnessed the emergence of a world market in the coal industry that is very different from the localized and specialized markets that existed prior to the 1970s. In addition, there have been three distinct periods for the dramatic oil price changes after World War II: falling prices from the 1940s for approximately 20 years until the late 1960s, sharply rising prices in the 1970s, and the resumption of declining prices in the 1980s and 1990s in the coal price market [9]. Our contributions in this study are as follows. First, the previous studies on energy price fluctuations only focus on testing for bubbles and the determinants for a single commodity’s bubbles but not consider the transmission mechanism within different energy commodities prices [10,11]. Our research selects the spot and the futures prices of several energy commodities, as we expand the scope of our study. Indeed, compared with previous studies, we pay more attention to the inner transmission mechanism over different types of energy price bubbles, and we further analyze the internal channels of the price-bubble transmission mechanism. Second, most studies have proved that the energy price has nonlinear characteristics (i.e., [12,13].2 We apply a newly nonlinear bubble detection method that can be applied to data at any frequency. In addition, this approach is able to account for a nonlinear structure and break in the course of investigating whether multiple bubbles exist. Furthermore, the Phillips, Shi, and Yu (PSY hereafter, 2013) method is a formal statistical technique for testing the existence of bubbles, whereas the other estimation approaches (i.e., the fundamental model approach and the cluster analysis approach) rely on subjective judgments of the deviations either from the fundamentals or from the moderate states. To the best of our knowledge, our study is the first paper to employ a right-tailed unit-root test to analyze energy price bubbles and the bubble transfer mechanism in several energy markets. Third, the volatility of energy prices has a significant impact on macroeconomic variables such as the consumer price index and economic growth. The analysis of energy price bubbles is suitable for understanding the energy price sensitivity, which is beneficial to investors’ scientific decisions in the face of multiple choices. The paper argues that the effective market hypothesis may not always hold in the markets. Our results support the hypothesis that the recent energy price run-up was amplified by speculative behavior during a bubble-like expansion. In addition, the gasoline spot price is volatile, and investors in gasoline markets have more speculative opportunities than investors in coal or oil markets. Moreover, it also presents hedge benefits in oil spot and futures markets, but there has not necessarily been any relationship between spot and future gasoline markets. In other words, the coal-gasoline price relation as well as the coal-crude oil price relation is a weak linkage, and the existent bubbles influence the interactions between oil and gasoline markets. The remainder of this empirical study is organized as follows: Section 2 gives a literature review, Section 3 describes the hypotheses and methodology of the recursive unit-root tests, Section 4 presents the data we used, and Section 5 discusses the findings. Finally, Section 6 shows our conclusions and policy implications.

2 Kyrtsou et al. [13] find evidence consistent with nonlinear dependencies in the energy market, suggesting that a successful nonlinear modeling of energy prices would produce richer energy market fluctuations than linear time series models. Kisswani and Nusair [12] test for nonlinearity and find significant evidence of it in oil prices.

277

2. Literature review The previous studies considering the commodity price bubble focus on testing for the existence of market bubbles or the time span of bubble periods [10]. Some scholars focus on comparing spot prices and futures prices in energy commodity markets [10,11]. Ghoshray and Johnson [14] show that the trend for energy prices is not well represented by a single positive or negative trend. The variability of the trend suggests that forecasting energy prices should not typically occur with a single trend. Nomikos and Andriosopoulos [15] indicate the presence of a “leverage effect” for West Texas Intermediate (WTI), Heating Oil and the Heating OilWTI crack spread, whereas the remaining energy markets exhibit an “inverse leverage” effect. Tsvetanov et al. [10] demonstrate that all of the pertinent series exhibited periods of bubble behavior that ended in late 2008. Moreover, dating algorithms have established that bubbles in longer-dated contracts start much earlier and are more persistent than bubbles in the spot contracts in the crude oil market. Narayan et al. [16] find dynamic trading strategies that suggest that all commodities are profitable, and the profits depend on structural breaks. The 2008 global financial crisis is marked as a period in which commodity profits were the weakest investment [17]. attribute this investment behavior to the specifics of commodity futures: although they may be predictable locally in the short term, they revert to their fundamental prices in the long term. A few researchers find a certain relationship between the two energy commodity prices or the links between the energy market price and other markets, such as the stock price [18]. Chen et al. [19] find that retail gasoline prices respond asymmetrically to crude oil price changes not only in short-run and long-run adjustments but also across the spot and future markets [20]. argues that there is a robust negative relationship between the oil price volatility and the asymmetry of the gasoline price. Miller and Ratti [21] support a conjectural change in the relationship between the real oil price and real stock prices in the last decade compared to earlier years, which may suggest the presence of several stock market bubbles and/or oil price bubbles since 2000. It really exits many bubble detection methods in the literature. One of the most common methods is the present value model and it shows that the price of an asset is discounted by the sum of future incomes if it has no bubble condition. It proves that the price of a financial asset is the present value of the future incomes according to the optimization approach without rational bubble and arbitrage conditions [22], which is the fundamental value of the price of an asset. If the investor pays more fundamental value of the price of the asset, the rational bubble has produced. At that time, the asset price could be divided into the fundamental value and the bubble component. Shiller [23] proposes the variance bounds test for bubble and it means that he variance of the asset price is more than the variance of the fundamental value under the rational bubble conditions. However, Yiu et al. [24] find bubbles from the test can be ruled out by other reasonable factors. West [25] offers the two-step test proposed under the equilibrium analysis of asset prices. Comparing to the underlying equilibrium model, if the linear model is no difference with it and the result means there is no bubble. While if there exits differences between the linear model and the underlying equilibrium model, it shows the bubble appearance. This method also depends on the strength of the equilibrium model and may reject the hypothesis due to a model misspecification [24]. Diba and Grossman [26] show that the differences between the asset price and the fundamental price are bubble detection and highlight the unit-root test is the appropriate approach for bubble detection. The fundamental model approach [27] requires the correctly choosing fundamental variables because the bubble

278

T.-Y. Liu, C.-C. Lee / Energy 150 (2018) 276e288

represented by the unexplained residual part. As for the cluster analysis approach [28], the biggest problem is that the method is not robust due to the initial allocations and data updates. Overall, the above two approaches need quarterly frequency fundamental variables of the data. Compared to the previous studies, most of researches choose the monthly data [10] and just focus on the single futures or spot markets [29,15]. Even some researches have focused on futures and spot markets but just tends to consider one or two energy markets [30]. Those studies have some limitations. First, the use of low frequency data in energy market cannot accurately analyze the specific time of price fluctuation. Because prices in energy markets fluctuate rapidly, the study of monthly data ignores many specific time and policy shocks especially for the investment in the short run. Second, the energy futures and the spot market are closely linked, both of which cannot be ignored. Due to energy futures can be converted into spot by delivery, for rational investors, the hedging between futures market and spot market are the most reasonable investment strategies. Therefore, it is necessary to pay attention to both markets. Thirdly, most of the previous methods on the topic ignore the beginning and ending time of the energy price bubble, which is the most important issues for the investors and policy makers. Also, compared with the previous methods, our study has two advantages as follows. One is that the method is most appropriate for a practical implementation with a time series, and it delivers a consistent date-stamping strategy for the origination and termination of multiple explosive periods. On the other hand, the method has no effected by the data frequency. It has more power than other methods and doesn’t consider the data frequency. Also, the judgment of the structural breaks from the nonlinear view relies on objective statistical analysis but not the traditional dummy variable setting. The simulation of the critical value has been affected by the bootstrap approach. 3. Hypotheses development The efficient market hypothesis (EMH) proposed by Fama [31]; it means that securities markets are efficient in reflecting information about individual stocks and about the stock market as a whole. When information arises, the news spreads quickly and is rapidly incorporated into the prices of securities. Thus, it does not allow arbitragers to yield positive expected net profit. Although many researchers have focus on the efficient market hypothesis [32,33], little literature is based on testing for energy-market price bubbles. It shows that if a market price bubble exists, the price fluctuation is beyond the normal level, which can only verify the failure of the efficient market hypothesis. In this paper, we test the efficient market hypothesis of the energy market by price bubble analysis and simultaneously consider the performances of spot and futures energy markets. Phillips et al. [34] develop the forward recursive test procedure. Their method implemented a right-side ADF test and a sup test, which can be described as follows. According to the ADF test for a unit root against the alternative of an explosive root (the righttailed), we estimate the following autoregressive specification by least squares.

pi ¼ ep þ fpi1 þ

N X



bN Dpi1 þ εp;i ; εp;i  NID 0; g2p



(1)

i¼2

Where pi represents the energy commodity price; ep is a constant, and εp;i is the error term. For a given value of the lag

parameter N,3 NID denotes independent and normal distribution. The unit root null hypothesis is H0: 4 ¼ 1 and the right-tailed alternative hypothesis is H1: 4 > 1. In forward recursive regressions, Equation (1) is estimated repeatedly, using subsets of the sample data incremented by one observation at each pass. If the first regression involves k0 ¼ ½nw0  observations for some fraction k0 of the total sample, [∙] signifies the integer part of its argument; subsequent regressions employ this originating data set supplemented by successive observations giving a sample of size k ¼ ½nw forw0  w  1. Denoting the corresponding t-statistic by ADFs, ADF1 corresponds to the full sample. Under the null hypothesis we have

Z

w

MdM ADFw 00 0 11=2 Z w 2 @ M A

(2)

0

And

Z

w

MdM sup ADFw 0 sup

w2½w0 ;1

w2½w0 ;1

0 Z @

0 w

(3)

11=2 M2 A

0

where M is the standard Brownian motion. A comparison of suprADFr with the right-tailed critical values Rw MdM from sup R 0w 1=2 makes it possible to test for a unit root against w2½w0 ;1 ð

0

M2 Þ

explosiveness. However, this testing procedure cannot date stamp the emergence or collapse of exuberance. To locate the origin and the conclusion of exuberance, one can match the time series of the recursive test statistic ADFr withw2½w0 ; 1, against the right-tailed critical values of the asymptotic distribution of the standard Dickey-Fuller t-statistic [35,36]. In particular, if we is the origination date and wf is the collapse date of the explosive behavior in the data, the construct estimates of these dates as follows: ∧

we ¼ inf

qw0

n n o ∧ o adf adf q:ADFq > van ðqÞ ; wf ¼ inf q:ADFq < van ðqÞ ∧

qwe

(4) where vadf an ðqÞis the right-side critical value of ADFq, corresponding to a significance level of an . In practice, it is conventional to set the significance level in the 1%e5% range. To achieve consistent esti∧

∧

mation of the date stampsfwe ; wf g, the significance level an needs to approach zero asymptotically, and correspondingly vadf an ðqÞmust diverge to infinity to eliminate the type I error asn/∞. We let an depend on n in the above formulae. In the long period data, it usually owns multiple bubbles, employing the SADF test may suffer from reduced power and fail to bubbles test. To overcome this weakness, Phillips et al. [34] provide the generalized sup ADF (GSADF) test which extends the sample sequence to a broader and more flexible range compared to the SADF test [35,36]. Suppose the regression sample starts from the w1 fraction of the total sample and ends at fraction w2; where w2 ¼ w1 þ wk and wk is the fraction of the sample size in the

3 We use significance tests to determine the lag order N, as suggested in Campbell and Perron [54].

T.-Y. Liu, C.-C. Lee / Energy 150 (2018) 276e288

regression. In addition to expanding the sample window wk, the GSADF test allows the sample starting point w1 to vary within its feasible range from 0 to 1- wk. The regression starts from the first observation when w1 ¼ 0 and w1 ¼1- wk; the regression sample covers the last observation. The respective ADF statistic is denoted wk byADFw . The GSADF statistic can be defined to be the largest ADF 1 statistic over the feasible ranges between wk and w1, and we denote this statistic by GSADF. That is,

( GSADF ¼

) wk ADFw 1

sup

sup

wk 2½w0 ;w1 

w1 2½0;1wk 

(5)

Under the null hypothesis that the true process is a random walk without drift, the asymptotic distribution of the GSADF statistic is

279

1 (Dollars per Gallon) with data from March 03, 2006 to March 28, 2014.5 The data source for UK Brent, WTI and Australian thermal coal is the International Financial Statistics Database published by the International Monetary Fund. Gasoline data are taken from the Energy Information Administration and Quandl Sever. The spot, free-on-board price is measured in US dollars per barrel, gasoline is given in the U.S. wellhead price in US dollars per thousand cubic feet and coal prices are in US dollars. Table 1 shows the basic statistical characteristics. Except for the coal market, all of the other market-price data are weekly data. According to the Jarque-Bera test shown in Table 1, not all of the data obey the normal distribution, and the WTI Oil Spot Price’s standard deviation is the largest in the energy market. Thus, the employed series in both quantity and prices are generated by a

8 9 2 3 > > > > Z w2 Z w2 > > > > 1 5 > > > > 4 w w MdM  MðrÞdr½Mðw Þ  Mðw Þ  > > k k 2 1 > > > > 2 w1 w1 < = sup

sup

> > > > > :

wk 2½w0 ;1 w1 2½0;1wk  > > w2 ¼w1 þwk > >

8 2 32 91=2 Z w2 < Z w2 = 1=2 M 2 dr  4 Mdr5 wk wk : ; w1 w1

The Wiener process has independent increments with distributionMðw2 Þ  Mðw1 Þ  Nð0; wk Þ. The asymptotic critical values are obtained by numerical simulations, for which the number of replications is 2000.4

4. Data The price series we examine correspond to oil, gasoline, and coal, and we use monthly or weekly observations over the period from January 1970 to March 2014 in spot and futures markets. The coal price is the monthly spot price from January 1970 to February 2014. The oil spot market includes the Weekly Europe Brent Spot Price FOB (Dollars per Barrel) with data from May 15, 1987 to March 28, 2014, and the Weekly Cushing, OK WTI Spot Price FOB (Dollars per Barrel) with data from January 03, 1986 to March 28, 2014. The oil futures market includes the Weekly Cushing, OK Crude Oil Future Contract 1 (Dollars per Barrel) with data from April 08, 1983 to March 28, 2014. The gasoline spot market includes the Weekly New York Harbor Conventional Gasoline Regular Spot Price FOB (Dollars per Gallon) and the Weekly U.S. Gulf Coast Conventional Gasoline Regular Spot Price FOB (Dollars per Gallon) with data from June 06, 1986 to March 28, 2014, the Weekly U.S. Regular Conventional Retail Gasoline Prices (Dollars per Gallon) with data from January 21, 1991 to March 28, 2014, the Weekly New York Regular Conventional Retail Gasoline Prices (Dollars per Gallon) with data from June 05, 2000 to March 28, 2014, and the Weekly Los Angeles Reformulated RBOB Regular Gasoline Spot Price (Dollars per Gallon) with data from September 12, 2003 to March 28, 2014. In addition, the gasoline futures market includes the Weekly New York Harbor Reformulated RBOB Regular Gasoline Future Contract

4 We use a bootstrap methodology to compute the finite sample distributions of the recently proposed tests. Monte-Carlo simulations indicate that the bootstrap methodology works well and allows us to identify explosive processes and collapsing bubbles. Compared to the Phillips (2013) ADF test, we also focus the bubbles’ relevance to the energy market.

> > > > > > > > > ;

(6)

nonlinear mechanism. The results show that each individual series exhibits general nonlinear serial dependence as well as nonlinearity in the mean, kurtosis, and skewness functions. 5. Empirical results We investigate whether the indexes displayed episodes of explosive behavior by the SADF and GSADF statistic developed by Phillips et al. [34]. Table 2 shows the estimated SADF and GSADF statistics along with respective sample critical values. As can be seen in Table 2, the null hypothesis of no bubble is strongly rejected by the GSADF tests in different energy markets except for the Los Angeles Reformulated RBOB Regular Gasoline Spot Price at the 1% significance level. In addition, our work on the GSADF statistics of the New York Harbor Reformulated RBOB Regular Gasoline Future Contract 1 market price rejects the null hypothesis at the 10% significance level. Phillips et al. [34] demonstrate that the moving sample GSADF diagnostic outperforms the SADF test based on an expanding sample size in detecting explosive behavior in multiple bubble episodes and seldom gives false alarms, even in relatively modest sample sizes; the main reason is that the GSADF test covers more subsamples of the data. Based on this argument, we can conclude that there is evidence of multiple bubbles in most energy markets. Hence, only the New York Harbor Reformulated RBOB Regular Gasoline Future Contract 1 market is in accordance with the efficient market hypothesis in all of the studied periods. To locate specific bubble periods, we compare the backward SADF statistic sequence with the 95% SADF critical value sequence obtained from Monte Carlo simulations with 2000 replications.

5 The abbreviations of the Coal Price, the Europe Brent Oil Spot Price FOB, the Cushing, OK WTI Oil Spot Price FOB, the Cushing, OK Crude Oil Future Contract 1, the U.S. Gulf Coast Conventional Gasoline Regular, the U.S. Regular Conventional Retail Gasoline Prices, the New York Regular Conventional Retail Gasoline Prices, the New York Harbor Conventional Gasoline Regular Spot Price FOB, the Los Angeles Reformulated RBOB Regular Gasoline Spot Price and the New York Harbor Reformulated RBOB Regular Gasoline Future Contract 1 are, respectively, CS, EBOS, WTIOS, CROF, GCGS, USGS, NYGS, NYHGS, LAGS and NYGF.

280

T.-Y. Liu, C.-C. Lee / Energy 150 (2018) 276e288

Table 1 Summary statistics of energy prices. Energy market

Mean

Std.

Skewness

Kurtosis

J.-B.

Number

Abbreviation

Period

Data Type

Coal Price

42.295

27.488

1.811

6.650

584.005

530

CS

Monthly

Europe Brent Oil Spot Price FOB Cushing, OK WTI Oil Spot Price FOB Cushing, OK Crude Oil Future Contract 1 U.S. Gulf Coast Conventional Gasoline Regular Spot Price FOB U.S. Regular Conventional Retail Gasoline Prices New York Regular Conventional Retail Gasoline Prices New York Harbor Conventional Gasoline Regular Spot Price FOB Los Angeles Reformulated RBOB Regular Gasoline Spot Price New York Harbor Reformulated RBOB Regular Gasoline Future Contract 1

43.199

34.176

1.090

2.769

280.671

1403

EBOS

Jan 1970Feb 2014 May15,1987eMar 28, 2014

Weekly

41.374

30.370

1.064

2.804

280.592

1474

WTIOS

Jan03,1986eMar 28, 2014

Weekly

40.312

29.248

1.193

3.152

385.127

1617

CROF

Apr08,1983eMar 28, 2014

Weekly

1.155

0.836

1.042

2.675

269.231

1452

GCGS

Jun06,1986eMar 28, 2014

Weekly

1.918

0.924

0.735

2.072

152.531

1211

USGS

Jan21,1991eMar 28, 2014

Weekly

2.638

0.885

0.057

1.669

53.704

722

NYGS

Jun 05, 2000eMar 28, 2014

Weekly

1.178

0.849

1.052

2.675

274.037

1452

NYHGS

Jun06,1986eMar 28, 2014

Weekly

2.277

0.656

0.082

2.158

16.912

551

LAGS

Sep12,2003eMar 28, 2014

Weekly

2.383

0.565

0.329

2.446

13.027

422

NYGF

Mar03,2006eMar 28, 2014

Weekly

Notes: Std. denotes the standard deviation. J.-B. denotes the Jarque-Bera test for normality.

Table 2 The SADF test and the GSADF test applied to Energy Price. Energy Market

SADF

SADF- Critical values 90%

95%

99%

CS EBOS WTIOS CROF GCGS USGS NYGS NYHGS LAGS NYGF

7.700*** 5.291*** 5.376*** 5.829*** 2.692*** 5.217*** 3.903*** 3.064*** 0.622 0.838

1.136 1.167 1.211 1.192 1.210 1.198 1.142 1.174 1.164 1.123

1.470 1.428 1.504 1.479 1.482 1.423 1.472 1.450 1.459 1.398

1.989 2.020 2.008 2.088 1.978 1.933 2.050 2.118 2.007 1.982

GSADF

GSADF -Critical values 90%

95%

99%

9.054*** 5.320*** 5.376*** 5.829*** 2.959*** 5.217*** 4.013*** 3.351*** 1.493 2.054*

1.926 1.971 2.005 1.946 1.950 1.947 1.966 1.979 1.964 1.979

2.185 2.259 2.230 2.147 2.167 2.152 2.251 2.207 2.225 2.221

2.706 2.718 2.637 2.630 2.569 2.592 2.777 2.720 2.681 2.734

Notes: Critical values of both tests are obtained from Monte Carlo simulation with 2000 replications. The smallest window has 10% of observations. “* (**/***)” signifies significance at the 10% (5%/1%) level.

Figs. 1-9 display the results for the date-stamping strategy for each energy markets.6 There exist significant bubbles in the research period. The initial size of the window for each time series data in this paper is 10% of the series. The GSADF test covers more subsamples of the data and has a greater window of flexibility. We find both the number of bubbles and the start-end time period of the bubbles during the sample periods. Overall, the average number of bubbles for ten energy markets is five, and the USGS market has the largest quantity of bubbles with ten. In contrast, the LAGS market experienced no bubbles during the sample period. Among the other energy markets, two markets have two bubble periods, one market has three bubble periods, one market has four bubble periods, one market has five bubble periods, and three markets have six bubble periods. In totality, the result indicates that the quantities of the NYHGS market’s bubbles are consistently more than the corresponding quantities in the NYGF market, and the quantities of the USGS’s

6 The Los Angeles Reformulated RBOB Regular Gasoline Spot Price market has no bubbles during the research periods by the test. Therefore, we don’t show the figure of the market.

bubbles are more than the corresponding quantities for the crude oil and coal markets in the U.S. Fundamentally, the volatility of energy prices depends on changes in the supply and demand of energy markets. In the 1970s, oil prices rose in the oil crisis, mainly on the supply side, which was affected by production cuts and wars among oil exporters. In the coal market, the oil crisis brought the coal industry back to development and energy demand grew rapidly. The 1980s was the fastest growing period for the coal industry. In 1990, world coal output reached 4716.7 million tons, the highest level in history, up 59.4% since 1980 [37]. Strengthening coal demand since 2003 has encouraged new investment in the mining sector. In 2007 the demand for coal has outstripped its supply, and price bubbles are inevitable [38]. In the long-term, the Chinese demand for coal could have large impacts on the global coal market since 2003 [39]. In Oil market, taking WTIO for example, prices rise sharply in 2007e2008 because ongoing growth in Chinese oil demand runs into a sudden and unexpected halt to a decade long increase in nonOPEC production. This caused a loss of OPEC spare capacity because increased demand for OPEC production runs ahead of increases in OPEC capacity which is reinforced by speculative expectations [40]. Oil use actually grew much faster than this trend during 2001e2005. Between 2004 and 2010, capacity to produce oil could grows by 16 million barrels a day-from 85 million barrels per day to 101 million barrels a day-a 20% increase. Such growth over the next few years would relieve the current pressure on supply and demand. Also, speculation could have edged producers into the discovery that small production declines could increase current revenues and may be in their long run interests as well. And the strong demand may have moved us into a regime in which scarcity rents, are now an important permanent factor in the price of oil [41]. During 2005 to 2008, the world oil market experiences the growing demand and stagnant supply, global economic growth was quite impressive and world oil consumption grew 5 mb/d over this period, or 3% per year. The strong demand pressures were the key reason for the steady increase in the price of oil over this period, though there was initially enough excess capacity to keep production growing along with demand. While the production did not continue to grow after 2005. Ongoing instability in places like Iraq

T.-Y. Liu, C.-C. Lee / Energy 150 (2018) 276e288

281

Table 3 PSY (2013) unit-root test and bubble periods of energy market.

CS EBOS WTIOS CROF GCGS USGS NYGS NYHGS LAGS NYGF

Bubble1

Bubble2

Bubble3

Bubble4

Bubble5

Bubble6

1974.51974.7 1990.9.71990.10.19 1990.9.211990.10.19 1986.5.91986.6.13 1990.8.241990.8.30 1995.6.121995.6.18 2001.10.222002.2.25 1990.8.241990.8.31 e 2008.5.232008.7.4

1980.41981.11 2005.8.122005.9.19 2005.8.262005.9.2 1990.9.211990.10.19 2005.9.22005.9.8 1998.12.71999.3.8 2003.2.102003.3.24 2005.9.22005.9.9 e 2008.11.142008.12.26

1993.21993.3 2006.4.212006.5.12 2007.10.192008.1.18 2005.8.122005.9.2 2005.9.302005.10.6 2000.2.212000.4.10 2005.8.152005.9.12

2004.22004.10 2006.6.302006.8.11 2008.2.152008.9.5 2006.4.212006.4.28

2007.102008.9 2007.11.92008.1.18

2008.2.152008.8.29

2006.6.72006.6.14

2007.10.192008.9.5

2000.5.152000.7.10 2008.4.282008.7.28

2005.3.282005.5.2 2008.12.12009.1.19

e 2011.4.82011.4.29

e

e

and Nigeria were a contributing factor. Another is that several of the oil fields that had helped sustain earlier production gains reached maturity with relatively rapid decline rates [41]. In the gasoline market, the high prices of 1975e1980 were largely supply driven, while the high prices of 2001e2006 were demand driven; the endogeneity of prices and quantities would result in more elastic elasticity estimates in the 1970s and 1980s and less elastic estimates in the 2000s [42]. It explores the implications of a joint of the global market for crude oil and the U.S. market for gasoline. Crude oil is the main input in the production of motor gasoline, the retail price of gasoline will in addition be affected by shocks to the U.S. demand for gasoline as well as by shocks to the ability of U.S. refiners to process crude oil. In the long run, 54% of the variation in the real price of gasoline in the U.S. is

Bubble7

Bubble8

Bubble9

Bubble10

2005.6.112005.10.24 2011.2.282011.6.6

2006.4.172006.8.21

2007.5.142007.6.4

2008.4.72008.9.29

2011.4.112011.5.16

e

e

e

e

e

driven by oil-market specific demand shocks. Since 2008, it has shown that the bubble passes from the crude oil market to the gasoline market [43]. While, Bachmeier and Griffin [44] find no evidence of asymmetry in gasoline prices with oil price during 1985e1998 because rigidities at the retail gasoline price level may well turn out to be correct. The causes of each bubble for every commodity could be provided statistically as follows. As for the CS market, after the first oil shock in 1973, the world’s energy consumption shifted to coal, which was the fundamental cause of the rise in coal prices. It suggest that the effects of long run increases in future demand on prices in 1980 should be fairly small since the long run supply function for coal is quite elastic. Short run demand shocks may have increased prices in the short run much more [45]. In 1991, world

200

150 10 100

8 6

50

4 0

2 0 -2 1970

1975

1980

1985

1990

1995

2000

2005

The coal price The back SADF sequence The 95% critical values sequence Fig. 1. Date-stamping bubble periods in the coal price: the GSADF test.

2010

282

T.-Y. Liu, C.-C. Lee / Energy 150 (2018) 276e288

160 120 80 6 40 4 0 2 0 -2 1990

1995

2000

2005

2010

Weekly Europe Brent Spot Price FOB (Dollars per Barrel) The back SADF sequence The 95% critical values sequence Fig. 2. Date-stamping bubble periods in the Europe Brent Spot Price: the GSADF test.

160 120 80

6 4

40

2

0

0 -2 -4 1990

1995

2000

2005

2010

Weekly Cushing, OK WTI Spot Price FOB (Dollars per Barrel) The back SADF sequence The 95% critical values sequence Fig. 3. Date-stamping bubble periods in the WTI oil price: the GSADF test.

coal output was more than 4.38 billion tons, a net increase of twothirds from the early 1970s, but the world’s coal output in 1993 was 44.02 billion tons, which was positive growth compared with the previous two years, and a 0.5% increased year-on-year. The intercept break for coal is at June 2002, a time at which coal prices rose rapidly as supply was not able to meet growth in demand [14]. In 2004, coal consumption accounted for 27.2% of global energy

consumption and coal prices increased by 40e80% over the previous year. Oil prices continue to rise sharply from 2005 to 2007 have failed to make the GDP grown but forcing the world has increased the demand for coal, leading to an average annual growth rate of coal consumption rose. In the second half of 2008, when the financial crisis hit, coal prices fell [46]. In EBOS market, between September and October in 1990, oil

T.-Y. Liu, C.-C. Lee / Energy 150 (2018) 276e288

283

160 120 80

6

40

4 2

0

0 -2 -4 1985

1990

1995

2000

2005

2010

Weekly Cushing, OK Crude Oil Future Contract 1 (Dollars per Barrel) The back SADF sequence The 95% critical values sequence Fig. 4. Date-stamping bubble periods in the Cushing, OK Crude Oil Future Contract 1 price: the GSADF test.

4 3 2 4 1 2 0 0 -2 -4 1990

1995

2000

2005

2010

Weekly U.S. Gulf Coast Conventional Gasoline Regular Spot Price FOB (Dollars per Gallon) The back SADF sequence The 95% critical values sequence Fig. 5. Date-stamping bubble periods in the U.S. Gulf Coast Conventional Gasoline Regular Spot Price: the GSADF test.

prices rose sharply, breaking through $40 per barrel for the first time, but quickly slipping below $20 a barrel two months later. In August 2005, when Mexico was hit by hurricane katrina, international oil prices topped $70 per barrel for the first time. Oil price rose from around $58 a barrel, and have fluctuated for months at $70 a barrel, at a top of $77.05 a barrel between April and August in 2006. Starting in November 2007, oil prices began to spurt upwards, with an extremely rare increase in momentum and peaked in July 2008. On September 12, 2007, the crude oil broke above $80 for the first time, and then continued to accelerate. The crude oil prices soared to $141.07 per barrel in July 2008. In WTIOS market,

Affected by the war between Iraq and Kuwait, WTI prices rose to $33.51 per barrel increased by 30% in September 1990. In August 2005, when Mexico experienced hurricane Katrina and hurricane Rita, international oil prices soared, and WTI also broke through $70 per barrel in August. For background it should be noted that in the monthly data, spot prices per barrel for WTI are $58.14 per barrel in January 2007 and $140 in June 2008 [18]. In CROF market, the price of crude oil rebounded after falling in May 1986, rising from 11.96 in March of that year to 15.75 Dollars per Barrel in May. In the first half of 2006, international oil prices climbed rapidly. Oil futures were trading at $72.14 per barrel in May

284

T.-Y. Liu, C.-C. Lee / Energy 150 (2018) 276e288

5 4 3 6 2 4 1 2 0 0 -2 -4 1995

2000

2005

2010

Weekly U.S. Regular Conventional Retail Gasoline Prices (Dollars per Gallon) The back SADF sequence The 95% critical values sequence Fig. 6. Date-stamping bubble periods in the U.S. Regular Conventional Retail Gasoline Price: the GSADF test.

5 4 3 5 2

4 3

1

2 1 0 -1 2000

2002

2004

2006

2008

2010

2012

Weekly New York Regular Conventional Retail Gasoline Prices (Dollars per Gallon) The back SADF sequence The 95% critical values sequence Fig. 7. Date-stamping bubble periods in the New York Regular Conventional Retail Gasoline Price: the GSADF test.

2006, up 44% to the previous year. From March of 2008 to July of 2008, during which time the price of crude oil exploded from slightly over $100 a barrel to almost $150 a barrel [47]. The reasons for bubbles during 1990.9.21e1990.10.19 and 2005.8.12e2005.9.2 are the same with the spot market EBOS or WTIOS. As for the GCGS market, the gulf war broke out in August 1990, with gasoline prices at $1.009 per barrel at the end of August, which has increased by 50% on year-on-year basis. The price of bubble is 25% higher than that of the previous period. Since 2005 when hurricane katrina hit the gulf coast, gasoline prices soaring to $2.77 a barrel (Sept. 30), an

increase of 51.04% compared to the same period last month. On October 14, it fell back into the middle of August. In USGS market, during the whole 1995, the gasoline price has peaked in June. The average number of gasoline price is $1.19 per barrel for the bubble which has increased 9.1% compared to the April 1995. During December 1998 to March 1999, the gasoline price in March 1999 is 0.9 while before that the price has decreased 26.5% since April 1996. In 2000, the international crude oil prices had a short period of time to above $30 per barrel, which has an impact on gasoline prices. Gasoline prices rose 19.11% from

T.-Y. Liu, C.-C. Lee / Energy 150 (2018) 276e288

285

4 3 2 4 1 2 0 0 -2 -4 1990

1995

2000

2005

2010

Weekly New York Harbor Conventional Gasoline Regular Spot Price FOB (Dollars per Gallon) The back SADF sequence The 95% values sequence Fig. 8. Date-stamping bubble periods in the Weekly New York Harbor Conventional Gasoline Regular Spot Price FOB: the GSADF test.

3.6 3.2 2.8 2.4 2.0 1.6

3

1.2 2

0.8

1 0 -1 -2 2006

2007

2008

2009

2010

2011

2012

2013

Weekly New York Harbor Reformulated RBOB Regular Gasoline Future Contract 1 The back SADF sequence The 95% ctitical values sequence Fig. 9. Date-stamping bubble periods in the New York Harbor Reformulated RBOB Regular Gasoline Future Contract 1 Price: the GSADF test.

February to July 2000, which is the first significant increase in the price of gasoline since 1991. Between March and May in 2005, the price of gasoline rose slightly and then increased rapidly in the following six months to October. Among them, it has been more than $3 per barrel in September 2005. It rose from $2.078 per barrel June 2005 to $3.037 per barrel in October, with a short-term rising of about 50%. After a small dip in the price of gasoline in the early stage, the new peak within April to August 2006 was basically in line with the previous bubble and In August 2006, it re-reached back $3.004 per barrel. From January 2007 to early June 2007, the price of gasoline rose 40.07% from $2.236 to $3.132 per barrel. From April to September 2008, international crude oil prices surged, and

U.S. gasoline prices hit a record high, reaching $4.054 per barrel in late August 2008. Gasoline prices went down and then rose to a new high from April to May 2011. It reached $3.907 per barrel on May 9, 2011, which increased by 38.4% compared to the previous periods. In NYGS market, the price of gasoline picked up slightly, and compared with the previous sharp decline, the price of gasoline showed a partial bubble period which has been rising by 11.45%. Between February and March 2003, the price of gasoline hit a local high of $1.776 per barrel since the end of 2002. During December 2008 to January 2009, the gasoline price has improved the amount of increase by 9.01%. Also, the reasons for the bubbles during

286

T.-Y. Liu, C.-C. Lee / Energy 150 (2018) 276e288 Table 4 Bubbles’ relevance to the energy market. Time (Year) 1986 1990 1995e2003 2005 2006 2007 2008 2011

CROF(1986.5.9) GCGS(1990.8.24)/EBOS(1990.9.7)/WTIOS(1990.9.21) and CROF(1990.9.21) USGS(2000.5.15)/NYGS(2001.10.22) USGS(2005.3.28)/EBOS(2005.8.12)/CROF(2005.8.12)/NYGS(2005.8.15) /WTIOS(2005.8.26)/NYHGS(2005.9.2)/GCGS(2005.9.30) USGS(2006.4.17)/CROF(2006.6.7)/EBOS(2006.6.30) USGS(2007.5.14)/WTIOS(2007.10.19) and CROF(2007.10.19)/EBOS(2007.11.9) EBOS(2008.2.15) and WTIOS(2008.2.15)/USGS(2008.4.7) /NYGS(2005.8.15)/NYGF(2008.11.14) NYGS(2011.2.28)/ NYGF(2011.4.8) / USGS(2011.4.11)

Note: The time in this Table is according to Table 3.

2005.8.15e2005.9.12, 2008.4.28e2008.7.28 and 2011.2.28e2011.6.6 are the same with other gasoline market such as USGS market. In NYHGS market, the reasons for the two bubbles are related to the same periods like NYGS. In August 24, 1990, the price of gasoline has peaked to $1.027/barrel. Also, In September 2, 2005, the price of gasoline has increased to 45.29% compared to the previous week. In NYGF market, in 2008, the supply was greater than demand of gasoline due to the impact of the international financial crisis. The price of international gasoline oil futures is higher than its intrinsic value. During 2008.5.23 to 2008.7.4, the price of gasoline has increased from $3.333/barrel to $3.534/barrel which is the first time to exceed $3/barrel since 2006. The average price is $1.058 per barrel during 2008.11.14 to 2008.12.26 which has increased by 9.07% compared to the price in January 2009. In April 8, 2011, the price of gasoline has been $3.202/barrel which has improved 4% compared to the previous week. In Table 4, we show the relevance of bubbles in oil and gasoline markets according to their appearance in the research period. Table 4 indicates the relationship in different energy market. According to the starting time of bubbles in each commodity market in Table 3. We find the results and quantifier evidence are as follows.7 Compared to the different energy markets (including the coal, oil, and gasoline markets), the last bubble of the coal and oil price markets is ended in 2008, whereas the last bubble of the gasoline markets is ended in 2011 during this period. The quantities of gasoline spot-market bubbles are more than the corresponding quantities for the coal or oil markets. Therefore, the gasoline spotmarket price is volatile, and investors in gasoline markets have more speculative opportunities compared to those in coal and oil markets. These persistent changes in the volatility can affect the gasoline and oil industrial producers and consumers risk management, and alter their incentives to invest in gasoline and oil inventories and facilities for production and transportation. Thus, understanding the energy price volatility is important for derivative valuation, hedging decisions, and decisions about investing in physical capital that is tied to the production or consumption of gasoline, oil, or coal [48]. also finds a statistically significant positive trend of volatility in energy markets, but not for crude oil. Since 2008, the change in the energy market structure has been attributed to several factors. In addition, as the resource basins in Europe and North America mature, production costs should rise and make long-distance trades to those regions more profitable [49]. Currently, increasing demand for gasoline has increased its price volatility. Compared to the spot market and futures market for oil and

7 Because the coal market has monthly data and little relevance to the others, we only choose to study the oil and gasoline markets.

gasoline, the starting and ending times of bubble periods of the EBOS contain the future markets’ bubble periods between the EBOS and the CROF. This result suggests that if the EBOS has a bubble, a CROF bubble may appear. The crude oil in the European spot market is greatly associated with the original futures market in the United States (CROF). The U.S. crude oil futures price rises due to the strong European market, the Brent crude oil quality is higher, and most of them have been exported from Europe to North America, which can also increase the future price of the CROF market. Several studies determine the prices for crude oils from different parts of the globe through cointegration [44]; [50]. The long-run relationship indicated by cointegration implies that the world oil market is unified. These results indicate that most analysts agree that there is unified world oil in the market, which enables us to assess upper and lower tail dependence and determine whether the world’s markets are somewhat regionalized or globalized. Compared to the WTIOS market, the CROF market contains more bubbles, and therefore, the stability of the market in the CROF is lower than WTIOS market; however, in another word, CROF market has more speculative opportunities. The WTI is considered a weak benchmark, in part because of declining output and bottlenecks in the transportation networks that reduce arbitrage opportunities relative to other crude oils. For example, the pipeline that transports the WTI’s crude can only move it north towards Cushing, and crude from the OK-WTI cannot be moved south to the Gulf of Mexico [50]. According to Table 4, it also has been bubbles influencing each other between the oil and gasoline markets. In most periods, gasoline market bubbles (especially the USGS market’s bubbles) have more appearance than other energy markets in both spot and future periods. In 2008, the spot prices for generic gasoline showed asymmetry in responding to crude oil price changes and asymmetry also appears in the responses of retail prices to wholesale price changes [51]. Between the EBOS and WTIOS spot crude oil markets, the appearances of bubbles have corresponded with each other except in 2006. In 2006, the EBOS had a bubble period; while WTIOS market did not. The capacity utilization of the U.S. refineries has decreased and the crude oil inventory has grown. At the same time, because it has been affected by the Middle East economic growth, Europe’s oil demand is increasing; hence, the expected price rises [52]. Before 2006, the EBOS market bubble induced a bubble in the WTIOS market. With respect to the USeEurope relationship, the actual oil price increases are more likely to have a greater impact on Europe than the U.S. due to the weakness of Europe [53]. In contrast, the WTIOS markets’ bubbles appear earlier than the EBOS markets’ bubbles due to the weakness of the American industrial economy and the development of the European economy. With respect to the gasoline spot and future markets, compared to the NYHGS, NYGS and NYGF three markets, the NYHGS market

T.-Y. Liu, C.-C. Lee / Energy 150 (2018) 276e288

had bubbles before 2006, whereas the NYGF market had a bubble after 2007. Thus, the bubbles in these markets did not appear simultaneously during 1986e2014. The NYGS market bubble periods were longer than the bubble period in the NYHGS market during 2008e2011, which implies that there has been bubble symbiosis between the NYGS and the NYHGS since 2008. The U.S. gasoline industry is dominated by a relatively small number of large producers whose prices are unregulated, which may have contributed to the divergence of gasoline prices from the economic fundamentals and helped produce high prices. Shapiro and Pham [30] find that during 2000e2006, the direction, magnitude, and seasonal variations of price increases in the current futures markets did not fully reflect the underlying economic fundamentals in the U.S. During 2006e2011, the gasoline futures price was much higher than these projected economic conditions. Hence, the gap between the wellhead prices and the futures prices has actually widened, suggesting that non-economic distortions have been even greater in the futures market than in the spot markets. The gasoline demand has become a focal point among both the users and the investors, especially in light of recent credit problems within the industry and the exposure of many hedge funds and other investors to energy-related investments. Also, the risk of war and the extreme weather condition can cause large movements in the values, which will be counted as bubbles. One famous example is the rising oil and gasoline prices in 2005, which is in turn caused by the disruptions by Hurricane Katrina in Gulf of Mexico oil production. Compared to the characteristics of bubbles, the GCGS market had the less bubbles during research period, and it has also had the shortest bubbles period. Compared with other energy markets, the speculation investment opportunity for investor in the GCGS market is difficult to practice since there only a very short bubble period in GCGS market. Still, the GCGS market price fluctuations have been small, which can improve the speed of the price back to its longterm equilibrium, as indicated by the EMH theory. This theory also finds that the USGS market, which represents the whole gasoline level in the U.S., has the most bubble quantizes, but it cannot include the bubbles in other small markets such as the NYHGS, NYGS, LAGS, and NYGF. Consequently, the gasoline markets in the U.S. have large fluctuations, and different markets are obviously distinctive. However, the GCGS markets have been closely linked with the NYHGS markets by bubbles covering period. Dynamic trading strategies suggest that although all commodities are profitable, the profits are hugely dependent on structural breaks. The 2008 global financial crisis marked a period in which the commodity profits were the weakest [16]. Considering the bubble relationship between different markets, the coal market was basically insulated from the oil and gasoline markets by the bubble before 2007. This fact suggests that the coal-gasoline price relation as well as the coal-crude oil price relation has a weak link. Gasoline prices were affected by crude oil, and gasoline and crude oil prices should be strongly associated because gasoline and crude oil are substitutes in energy consumption, and furthermore, they are complements and rivals in production. 6. Conclusions and policy implications In this study, we apply the newest recursive unit-root tests proposed by Phillips et al. [34] to identify the beginnings and the ends of potential speculative bubbles in the energy market, including the Brent and WTI crude oil prices, the U.S. regular gasoline prices, and Australian coal prices’ during 1970e2014. These results support the hypothesis that the recent energy price run-up was amplified by speculative behavior during a bubble-like expansion. In addition, we examine the EMH theory. Compared to

287

previous studies, the traditional bubble test method requires fundamental variables/indicators, and some of them are sensitive to the initial allocations and data updates; the new method is much more suitable for a practical implementation with time series and incorporates a consistent date-stamping strategy for the origination and termination of multiple bubbles. Simulations demonstrate that the test significantly improves the discriminatory power and leads to distinct power gains when multiple bubbles occur. The major advantage of this approach is that it allows one to account for nonlinear structures and break in the course of investigating the existence of multiple bubbles. The results are as follows: firstly, the energy price run-up was amplified by speculative behavior of the type found during a bubble-like expansion. Furthermore, the gasoline spot market price is volatile, and investors in gasoline markets have more speculative opportunities than those investors who in coal and oil markets. Although there are hedging benefits in oil spot and future markets, there are not necessarily any relationships between spot and future gasoline markets. In addition, coal and gasoline prices as well as the interactions between coal and crude oil have very weak links, and the existent bubbles influence the interactions between the oil and gasoline markets. Our results verify the efficient market hypothesis across various energy markets. The results have three important implications. First, market speculation crisis could be caused by energy-price bubbles. When different energy markets’ bubbles burst, the economic effect is significant and the volatility increases the market instability. Because national economic development mainly depends on traditional energy, energy bubbles will drive the prices of related industries, which could lead to bubbles bursting in these industries. Second, energy-market price volatility will affect the national economic structure. The energy consumption may be artificially high with a rising price bubble, which would indirectly hinder the transformation and upgrading of industrial structures. Furthermore, energy price bubbles will limit the production of basic energy, and hence, of related industries. Third, investors can choose appropriate long-term investments in the energy industry and do not need frequent changes for some energy markets, which can reduce the turnover rate and transaction costs. When the international energy prices are volatile, investors can adjust their portfolios’ scales with respect to the energy industry or its products without overreacting. Lastly, our work provides a reference for dealing with energy-market bubbles in investment. Acknowledgements We would like to thank Editor and three anonymous referees for their highly constructive comments. Tie-Ying Liu is grateful to the financial support from the National Social Science Fund under Grant [17CJL016]. Chien-Chiang Lee appreciates financial support from the Taiwan Ministry of Science and Technology (Project: 1062918-I-110-002). References [1] Lee CC, Lee JD. A panel data analysis of the demand for total energy and electricity in OECD countries. Energy J 2010;31(1):1e24. [2] Lee CC, Lee CC, Ning SL. Dynamic relationship of oil price shocks and country risks. Energy Econ 2017;66:571e81. [3] Atil A, Lahiani A, Nguyen DK. Asymmetric and nonlinear pass-through of crude oil prices to gasoline and natural gas prices. Energy Pol 2014;65: 567e73. [4] Brewer J, Nelson DM, Overstreet G. The economic significance of gasoline wholesale price volatility to retailers. Energy Econ 2014;43:274e83. [5] Zhang YJ, Wang ZY. Investigating the price discovery and risk transfer functions in the crude oil and gasoline futures markets: some empirical evidence. Appl Energy 2013;104:220e8. [6] Ji Q. System analysis approach for the identification of factors driving crude oil

288

T.-Y. Liu, C.-C. Lee / Energy 150 (2018) 276e288

prices. Comput Ind Eng 2012;63(3):615e25. [7] Sornette D, Woodard R, Zhou WX. The 2006e2008 oil bubble: evidence of speculation, and prediction. Phys Stat Mech Appl 2009;388(8):1571e6. [8] Dahl CA. Measuring global gasoline and diesel price and income elasticities. Energy Pol 2012;41:2e13. [9] Ellerman AD. The world price of coal. Energy Pol 1995;23(6):499e506. [10] Tsvetanov D, Coakley J, Kellard NM. Bubbling over! the behaviour of oil futures along the yield curve. J Empir Finance 2015;38b:516e33. [11] Wolff H. Value of time: speeding behavior and gasoline prices. J Environ Econ Manag 2014;67(1):71e88. [12] Kisswani KM, Nusair SA. Non-linearities in the dynamics of oil prices. Energy Econ 2013;36:341e53. [13] Kyrtsou C, Malliaris AG, Serletis A. Energy sector pricing: on the role of neglected nonlinearity. Energy Econ 2009;31(3):492e502. [14] Ghoshray A, Johnson B. Trends in world energy prices. Energy Econ 2010;32(5):1147e56. [15] Nomikos N, Andriosopoulos K. Modelling energy spot prices: empirical evidence from NYMEX. Energy Econ 2012;34(4):1153e69. [16] Narayan PK, Narayan S, Sharma SS. An analysis of commodity markets: what gain for investors? J Bank Finance 2013;37(10):3878e89. [17] Kristoufek L, Vosvrda M. Commodity futures and market efficiency. Energy Econ 2014;42:50e7. [18] Ratti RA, Vespignani JL. Why are crude oil prices high when global activity is weak? Econ Lett 2013;121(1):133e6. [19] Chen LH, Finney M, Lai KS. A threshold cointegration analysis of asymmetric price transmission from crude oil to gasoline prices. Econ Lett 2005;89(2): 233e9. [20] Radchenko S. Oil price volatility and the asymmetric response of gasoline prices to oil price increases and decreases. Energy Econ 2005;27(5):708e30. [21] Miller JI, Ratti RA. Crude oil and stock markets: stability, instability, and bubbles. Energy Econ 2009;31(4):559e68. [22] Blanchard O, Watson M. Bubbles, rational expectations, and financial markets. In: Wachter P, editor. Crises in the economic and financial structure. Lexington, MA: Lexington Books; 1982. p. 295e315. [23] Shiller R. Do stock prices move too much to be justified by subsequent changes in dividends? Am Econ Rev 1981;71:421e36. [24] Yiu MS, Yu J, Jin L. Detecting bubbles in Hong Kong residential property market. J Asian Econ 2013;28:115e24. [25] West KD. A specification test for speculative bubbles. Q J Econ 1987;102(3): 553e80. [26] Diba BT, Grossman HI. Explosive rational bubbles in stock prices? Am Econ Rev 1988;78(3):520e30. [27] Ahuja A, Porter N. Are house prices rising too fast in Hong Kong SAR? (December 2010). IMF Working Papers. 2010. p. 1e19. Available at: SSRN, https://ssrn.com/abstract¼1751385. [28] Leung F, Chow K, Genberg H. Long-term and short-term determinants of property prices in Hong Kong. 2008. HKMA Working Paper 15/2008. [29] Lien D, Yang L. Asymmetric effect of basis on dynamic futures hedging: empirical evidence from commodity markets. J Bank Finance 2008;32(2): 187e98. [30] Shapiro RJ, Pham ND. An analysis of spot and futures prices for natural gas: the roles of economic fundamentals, market structure, speculation, and manipulation. 2006. Working Paper. [31] Fama EF. Efficient capital market: a review of theory and empirical work.

J Finance 1970;25(2):383e417. [32] Lee CC, Tsong CC, Lee JF. Testing for the efficient market hypothesis in stock prices: international evidence from non-linear heterogeneous panels. Macroecon Dyn 2014;18(4):943e58. [33] Smith VL, Suchanek GL, Williams AW. Bubbles, crashes, and endogenous expectations in experimental spot asset markets. Econometrica 1988;56(5): 1119e51. [34] Phillips PCB, Shi SP, Yu J. Testing for multiple bubbles 1: historical episodes of exuberance and collapse in the S&P 500. Singapore Management University; 2013. Working Paper No. 04-2013. [35] Phillips PCB, Shi S, Yu J. Testing for multiple bubbles. Research Collection School Of Economics; 2011. Available at: http://ink.library.smu.edu.sg/soe_ research/1302. [36] Phillips PCB, Wu Y, Yu J. Explosive behavior in the 1990s NASDAQ: when did exuberance escalate asset values? Int Econ Rev 2011;52(1):201e26. [37] Pan HF, Lv WD, Zheng JM. The pattern and price change trend of global coal market. J Shanxi Univ 2008;31(4):83e9. [38] Rademacher M. Development and perspectives on supply and demand in the global hard coal market. Z Energiewirtschaft 2008;32(2):67e87. [39] Yang CJ, Xuan X, Jackson RB. China’s coal price disturbances: observations, explanations, and implications for global energy economies. Energy Pol 2012;51:720e7. [40] Kaufmann RK. The role of market fundamentals and speculation in recent price changes for crude oil. Energy Pol 2011;39(1):105e15. [41] Hamilton JD. Understanding crude oil prices. Energy J 2009;30(2):179e206. [42] Hughes JE, Knittel CR, Sperling D. Evidence of a shift in the short-run price elasticity of gasoline demand. Energy J 2006;29(1):113e34. [43] Kilian L. Explaining fluctuations in gasoline prices: a joint model of the global crude oil market and the US retail gasoline market. Energy J 2010;31(3): 87e112. [44] Bachmeier LJ, Griffin JM. New evidence on asymmetric gasoline price response. Rev Econ Stat 2003;85(3):772e6. [45] Joskow P. Price adjustment in long term contracts: the case of coal. Massachusetts institute of Technology (MIT), Department of Economics; 1987. [46] Zhang K, Liang H. Global coal consumption changes during 1999-2009 and enlightenment. Sino-Global Energy 2010;15(8):7e14. [47] Tokic D. Rational destabilizing speculation, positive feedback trading, and the oil bubble of 2008. Energy Pol 2011;39(4):2051e61. [48] Pindyck RS. Volatility in natural gas and oil markets. J Energy Dev 2004;30(1): 1e19. [49] Brito DL, Hartley PR. Expectations and the evolving world gas market. Energy J 2007;28(1):1e24. [50] Kaufmann RK, Ullman B. Oil prices, speculation, and fundamentals: interpreting causal relations among spot and futures prices. Energy Econ 2009;31(4):550e8. [51] Borenstein S, Cameron AC, Gilbert R. Do gasoline prices respond asymmetrically to crude oil price changes? Q J Econ 1997;112(1):305e39. [52] Zhu WT, Xu ZH. The cause and effect of upside down between WTI and Brent crude oil prices down. International Petroleum Economy 2011;19:8e15. ~ ado J, Pe rez de Gracia F. Do oil price shocks matter? Evidence for some [53] Cun European countries. Energy Econ 2003;25(2):137e54. [54] Campbell JY, Perron P. Pitfalls and opportunities: what macroeconomists should know about unit roots. In: NBER macroeconmics annual, 6. MIT Press; 1991. p. 141e220.