Dynamic effects of rising oil prices on consumer energy prices in Canada and the United States: Evidence from the last half a century

Energy Economics 45 (2014) 33–44

Contents lists available at ScienceDirect

Energy Economics journal homepage: www.elsevier.com/locate/eneco

Dynamic effects of rising oil prices on consumer energy prices in Canada and the United States: Evidence from the last half a century☆ Abbas Valadkhani ⁎ Swinburne Business School, Swinburne University of Technology, Hawthorn, VIC 3122, Australia

a r t i c l e

i n f o

Article history: Received 27 December 2013 Received in revised form 18 June 2014 Accepted 23 June 2014 Available online 5 July 2014 JEL classification: C51 D40 E21 E27 E31 E37

a b s t r a c t This paper examines the dynamic relationship between the price of crude oil and the CPI energy price sub-index in Canada and the U.S. using a Markov-regime switching model and the Bai–Perron sequential method. Since these two series are cointegrated during the sample period (January 1961–June 2013), a short-run dynamic model is thus estimated for each country in which all coefficients and the error-variance terms can freely switch over time between two values prevailing in Regimes 0 and 1. Previous studies indicate that the price of crude oil does not currently affect the aggregate CPI as much as it did in the 1970s. This finding is not disputed in this paper. However, the sequentially-determined break date as well as the time-varying regime-switching probabilities point to two new findings. First, the marginal effects of changes in oil price on consumer energy prices (not the aggregate CPI) have consistently increased and become more instantaneous for both countries after the Western U.S. Energy Crisis of 2000. Second, the speed of adjustment (proxied by different error-correction coefficients) has also risen, particularly for the U.S. Therefore, oil prices exert far more positive and immediate impacts on energy costs in the post- rather than pre-1999 periods. © 2014 Elsevier B.V. All rights reserved.

Keywords: Canada U.S. Crude oil price CPI inflation Energy costs

1. Introduction Oil prices can influence many vital economic indicators in an economy such as inflation, industrial production and stock market returns and volatility (Ali Ahmed and Wadud, 2011; Burbridge and Harrison, 1984; Filis, 2010; Hamilton, 1983; Kilian, 2008a, 2008b; Miller and Ratti, 2009; Ratti and Hasan, 2013; Ratti and Vespignani, 2013). There is a consensus among economists that changes in oil prices (particularly large variations) can exert significant influences on inflation (see inter alia Ajmera et al., 2012; Bachmeier and Cha, 2011; Goldberg, 1998; Nixon and Smith, 2012). For example, Nixon and Smith (2012) argue that producing an accurate inflation forecast without knowing the future path of oil and other commodity prices can be futile. They believe that when forecasting CPI inflation and GDP growth, the UK Monetary Policy Committee rigorously examines the slope of the oil future curve as it

☆ I wish to thank Professor Ang Beng Wah (the Editor) and three anonymous referees, whose useful comments considerably improved an earlier version of this article. The usual caveat applies. ⁎ Tel.: +61 3 9214 8791. E-mail address: [email protected].

http://dx.doi.org/10.1016/j.eneco.2014.06.015 0140-9883/© 2014 Elsevier B.V. All rights reserved.

contains predictive information for explaining the expected path of spot prices. Ratti and Vespignani (2013) attribute significant increases in real oil prices, global oil production, and global real aggregate demand to unanticipated rises in global real M2 particularly in the post- rather than pre-2005 period. In this comprehensive study they find that Brazil, China, India and Russia reinforce one another in terms of their liquidity effects on global real aggregate demand. On the other hand, Kilian (2008a) compares the impacts of exogenous shocks to oil supply on G7 economies in terms of their responses to output growth and inflation. Although such transitory disruptions do not exert sustained inflationary or deflationary pressure on any of these economies, inflation responses are mainly country-specific and typically peak after three to four quarters (Kilian, 2008a). He also asserts that exogenous shocks to oil supply can result in a fall in the real wage, higher short-term interest rates, and a depreciation of the local currency. Using monthly data (1986–2009), Ali Ahmed and Wadud (2011) examine the impact of oil price uncertainty on Malaysian macroeconomic activities and monetary responses by estimating a SVAR. They argue that the Malaysian central bank pursues an expansionary monetary policy in response to oil price uncertainty due to prolonged and dampening effects of oil price volatility shock on industrial production.

34

A. Valadkhani / Energy Economics 45 (2014) 33–44

Ajmera et al. (2012) analyse price movements of four commodities in the U.S. (namely crops, animal slaughter and processing, dairy, and oil and gas) and estimate price transmission from commodity prices to various sub-components of the consumer price indices. Out of these four commodities, they argue that only oil and gas prices have a significant impact on inflation both in the longer term as well as in specific years through influencing the three CPI sub-indices: transportation; motor fuel; and fuels and utilities (including electricity and natural gas). According to Ajmera et al. (2012, p.39) between 2003 and 2008 higher oil and gas prices were responsible for “approximately twothirds of the increase in transportation prices, about one-half of the increase in motor fuel prices, and roughly two-thirds of the increase in fuels and utilities prices.” Using the monthly Greek data during the period 1996 m1–2008 m6, Filis (2010) also studied the interplay among the cyclical components of the CPI, industrial production, stock market and oil prices within a multivariate VAR framework. His results also indicate that oil prices exert a positive effect on the CPI in the long run. A number of studies in the literature support this view that the recent effects of oil shocks on core aggregate inflation (not energy prices) are no longer as large and significant as they were back in the 1970s because producers have substituted away from oil (see inter alia Bachmeier and Cha, 2011; De Gregorio et al., 2007; Hooker, 2002; Katayama, 2013). For example, Bachmeier and Cha (2011) used a disaggregate monthly database (97 sectors) to examine the relationship between oil price shocks and U.S. CPI inflation. Their comprehensive database contains sectoral inflation data, energy intensity, labour intensity, and sensitivity to monetary policy. Comparing the 1973–1985 and 1986–2006 time periods, inter alia they posit “that about two-thirds of the reduced response of core inflation to oil shocks can be attributed to changes in energy intensity and one-third to monetary policy” (Bachmeier and Cha, 2011, p.1167). Oil and energy prices are typically more volatile than the prices of 95% of products sold by domestic U.S. producers (Regnier, 2007). Cologni and Manera (2008) examined the direct impacts of oil price shocks on output, prices and monetary variables within a structural VAR model for the G-7 countries. They found, inter alia, that for Canada and the U.S. the null hypothesis of an influence of oil prices on the inflation rate could not be rejected. Wang (2013) in his study of the G-7 countries used a logistic smooth transition model and argued that the threshold effects of rising oil prices on household consumption expenditures are larger than those of falling oil prices. He found that such influences arise from two channels: real balance effects and wealth transfer effects. On the same topic Venditti (2013) examined the effects of oil prices on weekly energy prices (gasoline and gasoil) in the U.S. and the four euro area countries (Germany, France, Italy and Spain) using nonlinear impulse response functions and forecast accuracy tests. According to Venditti (2013), there is some compelling evidence of asymmetries in the adjustment of retail prices for the U.S. but such evidence appears to be quite mixed for the euro area. There has been little research in measuring the direct influence of the price of crude oil on the energy price sub-index within the CPI basket. Canada and the U.S. are chosen in the present study because these two countries possess the highest per capita consumption of petroleum-based liquid fuel among all OECD countries (Knittel, 2012). While concurring with previous studies in that the positive effect of rising crude oil prices on the overall CPI has diminished over time, the same cannot be said in relation to the dynamic impacts of oil prices on the CPI energy sub-index. The empirical results of this study provide convincing evidence that oil prices can influence consumer energy prices both faster and on a larger scale in the post-1999 period than they did in the pre-1999 period. If producers had not efficiently substituted away from oil due to technological advancements, rising oil prices would have increased consumer energy bills (i.e. fuel, transport as well as utilities including electricity and natural gas) far more than what is currently observed.

The remainder of this paper is organised as follows: Section 2 discusses two modelling approaches (i.e. Markov-regime switching model and the Bai–Perron sequential approach) in capturing any break or switching effects of rising crude oil prices on the consumer energy price sub-index. Section 3 briefly outlines the sources and summary statistics of the data employed. Section 4 presents empirical results and the major findings of the study followed by some concluding remarks in Section 5. 2. Econometric methodology As an important production input crude oil can be refined into various types of fuels (including kerosene, gasoline and diesel) and other petroleum products. It is reasonable thus to assume that changes in the price of crude oil can have an appreciable effect on household energy bills. The production and consumption of other sources of energy such as natural gas and coal can also be indirectly influenced by the price of crude oil. One expects that the marginal effects of changes in the price of crude oil on the consumer price index vary due to (a) technological advancements in the use of this important production input; (b) environmental considerations and the use of other sources of energy. Given that oil prices can be influenced by a wide range of socioeconomic and political factors as well as natural catastrophes in an unpredictable manner, this paper assumes that the marginal effects are random, variable and “memoryless”. In other words, the marginal effects of changes in oil prices on consumer energy prices can vary from one state to another on a state space whereby the next state is dependent upon only the current state and not on the sequence of events that preceded it. A Markov chain process can capture these three attributes: randomness, time-variant and “memorylessness.” (Valadkhani, 2014). In addition, previous studies found that the effect of crude oil prices on the overall CPI has undergone significant changes since the 1970s. In order to capture any possible dynamic changes in the marginal effect of oil prices on the CPI energy price index, a hybrid Markov-switching model is adopted below in which both the marginal effects and the variance of error term are allowed to be regime variant. Let us assume that the energy price index (Et) within the CPI basket and the price of crude oil (POt) are both I(1) and they are also cointegrated in the long run. Due to possible endogeneity between these two variables, the Dynamic Least Square (DLS) method (Stock and Watson, 1993) can then be used to estimate the long-run coefficients (ηs) as follows: LnðEt Þ ¼ η0 þ η1 T t þ η2 LnðPOt Þ þ

þk X

  φ j ΔLn POt− j þ vt

ð1Þ

j¼−k

where Tt = the time trend variable; Ln = natural logarithm; and vt = the residual term. Based on the Schwarz criterion, the optimal k lags and leads of ΔLn(POt-j) are included in Eq. (1) to address the possibility of endogeneity problem raised by Kilian (2008b). The long-run elasticity of oil price (η2) is expected to be positive as higher crude oil prices lead to higher energy prices for consumers. According to the Engle–Granger theorem, the lagged residual term obtained from the long-run cointegrating vector can form an error correction (ECt-1) term in a short-run dynamic model. For simplicity let us first define:  Y t ¼ ΔLnðEt Þ ð2Þ X t ¼ ΔLnðPOt Þ One can then write the following conventional (non-switching) short-run dynamic model for the growth rate of the energy price index: Yt ¼ α þ

q X i¼0

q X βi X t−i þ γi Y t−i þ λEC t−1 þ ut

ð3Þ

i¼1

where βi is the short-run effect of a logarithmic change in the price of crude oil on the growth rate of the energy price index at time t-i; γi

A. Valadkhani / Energy Economics 45 (2014) 33–44

denotes the impact of the corresponding lagged dependent variable (a proxy for the level of inertia) at time t-i; λ is the feedback coefficient assigned to the error correction term, quantifying the extent to which any disequilibrium between actual Ln(Et) from its estimated long-run path is eliminated within each time period; and μt is a white noise req

q

i¼0

i¼1

sidual term. Theoretically, it is expected that ∑ βi N0, λ b 0 and ∑ γ i will be well below unity. In a conventional error correction model such as Eq. (3) it is typically assumed that the estimated short-run coefficients (i.e. α, βi,, γi and λ) as well as the variance of the residual (i.e. σ2u) all remain unchanged over time. However, when the sample period contains more than half a century of monthly data these assumptions can be unrealistic. In order to capture/test the possibility of a regime change in any of the above parameters, the following short-run dynamic switching model is thus proposed: EðY t jSt Þ ¼ ð1−St Þα 0 þ St α 1 q q q X X X þ ð1−St Þ β0i X t−i þ St β1i X t−i þ ð1−St Þ γ 0i Y t−i þ i¼0 i¼0 i¼1 q X þSt γ 1i Y t−i þ ð1−St Þλ0 EC t−1 þ St λ1 EC t−1 þ ut i¼1

ð4Þ

Q t ¼ EðY t jSt Þ þ ð1−St Þσ 0 ut þ St σ 1 ut

ð5Þ

where St is a discrete random variable which takes two possible values (0, 1), representing the state of the economy at time t. Unlike Eq. (3), all of the parameters of the above model (including the variance term) can switch over time between two possible values prevailing in each of the two regimes indicated by the subscripts 0 and 1. Eqs. (4) and (5) can thus provide a more complete account of the nonlinear-regime-dependent effects of the price of crude oil on the consumer energy price index by allowing all parameters to follow a Markov chain process. The dependent variable (Yt) at any point of time can respond differently to an equal change in Xt depending on whether or not the economy is in a particular state. The unobserved state variable in turn follows the first-order Markov switching process discussed in Hamilton (1989, 1996) whereby: 8 P ½S > > < t P ½St > > P ½St : P ½St

¼ 1jSt−1 ¼ 1jSt−1 ¼ 0jSt−1 ¼ 0jSt−1

¼ 1 ¼ p ¼ 0 ¼ 1−q ¼ 0 ¼ q ¼ 1 ¼ 1−p

ð6Þ

where p and q denote the fixed transition probabilities of being in regime 0 or 1, respectively assuming 0 b q b 1 and 0 b p b 1. In a process referred to as filtering the joint conditional probabilities of the present state and the transition probabilities are used to estimate the above Markov model. The joint probability of Qt and St is obtained in two steps. First, the conditional probability of being in a particular state such as st (given all available information at time t-1) is estimated by: P ðSt ¼ st jQ t−1 Þ ¼ P ðSt ¼ st jSt−1 ¼ st−1 Þ  P ðSt−1 ¼ st−1 jQ t−1 Þ

ð7Þ

Then by utilising the joint probability density distribution of Qt and St and all information available at time t, the probability of being in state st is iteratively revised to P(St = st|Qt) as follows: f ðQ t ; St ¼ st jQ t−1 Þ ¼ f ðQ t ; St ¼ st jQ t−1 Þ  P ðSt ¼ st jQ t−1 Þ

ð8Þ

Eq. (8) can also be written as: f ðQ t ; St ¼ st jQ t−1 Þ ¼ f ðQ t ;St ¼ st jQ t−1 Þ  P ðSt ¼ st jSt−1 ¼ st−1 Þ  P ðSt−1 ¼ st−1 jQ t−1 Þ

ð9Þ

35

Given that the density distribution of Qt is defined by: f ðQ t jQ t−1 Þ ¼ f ðQ t ; St ¼ 0jQ t−1 Þ þ f ðQ t ; St ¼ 1jQ t−1 Þ

ð10Þ

The updated joint probability of Qt and St is thus obtained as follows: P ðSt ¼ st jQ t Þ ¼

f ðQ t ; St ¼ st jQ t−1 Þ f ðQ t jQ t−1 Þ

ð11Þ

Assuming that the residual series is normally distributed,1 the parameters of the model Ω : {α0, α1, β0i, β0i, γ0i, γ1i, λ0, λ1, σ0, σ1} can be formulated based on the following maximum likelihood function: LðΩÞ ¼

T X 1 1 X X

f f ðQ t jSt ¼ st ; Q t−1 ; ΩÞ  P ðSt ¼ st jSt−1 ¼ st−1 Þ  P ðSt−1 ¼st−1 jQ t−1 Þg

t¼1 st ¼0 st−1 ¼0

ð12Þ Based on the Broyden, Fletcher, Goldfarb and Shanno (BFGS: Broyden, 1970; Fletcher, 1987) iterative procedure, Eq. (12) presents a full nonlinear normal mixture which can then be maximised with respect to each of the parameters of the model: i.e. {α0, α1, β0i, β0i, γ0i, γ1i, λ0, λ1, σ0, σ1}. The estimated local optima arising from Eq. (12) are unbounded but are consistent and asymptotically efficient. The variance and covariance of the estimators are estimated in EViews 8 utilising the degree-of-freedom corrected inverse of the negative of the observed Hessian. In order to enhance the estimated joint probabilities, Kim's (1994) smoothing process is also employed. One can also compare and contrast the estimated switching parameters in Eq. (4) with the endogenously-determined corresponding coefficients in Eq. (3) by employing the sequential-search method for break date proposed by Bai (1997) and Bai and Perron (1998). This iterative approach can detect m structural break dates resulting in m + 1 subperiod parameter estimates within the entire sample period. Therefore, all parameters in Eq. (3) are subject to a single structural (m = 1) or regime change but employing two totally different estimation methods. Similar to Eq. (4) two sets of parameter estimates are obtained within the sample period. The break date in this method is not known a priori and will be determined in an iterative procedure if it is found to be statistically significant (where α = 5%). The resulting break dates from the Bai and Perron sequential approach can then be compared to the smoothed regime-switching probabilities to see if any discernible comparison can be made over the same period. 3. Data sources and descriptive statistics The sample period in this study spans from January 1961 to June 2013 (containing 630 monthly observations). Time series data on the price of oil (West Texas, Spot, fob Midland Texas, $US/barrel) and various CPI sub-indices were sourced from Econdata (2013, IFS database). All seasonally adjusted sub-CPI indices (i.e. energy, food, non-food, non-energy and aggregate) are measured in local currency and then indexed to 2005 = 100. In order to display CPI indices (2005 = 100) and the price of oil ($US/barrel) in one single graph, the original series are normalised to have the same measurement scale. As can be seen from Fig. 1, the relationship between the price of crude oil and the aggregate CPI indices for both Canada and the U.S. does not appear to be strong, particularly in the post-1970 period. This observation is broadly consistent with the findings of earlier studies outlined in the previous section (i.e. Bachmeier and Cha, 2011; De Gregorio et al., 2007; Hooker, 2002). However, according to Fig. 2, the energy price indices of Canada and the U.S. closely follow the price of crude oil throughout the sample period, particularly since the post-2000 period. Switching to growth 1 With a large sample of T = 630 observations one should not be too concerned about this assumption as the estimated results will be asymptotically justified as the method of moments (Hayashi, 2000, Sections 2.7–2.10).

36

A. Valadkhani / Energy Economics 45 (2014) 33–44

5 4

Normalised data

3 2 1 0 -1

1961m10 1963m6 1965m2 1966m10 1968m6 1970m2 1971m10 1973m6 1975m2 1976m10 1978m6 1980m2 1981m10 1983m6 1985m2 1986m10 1988m6 1990m2 1991m10 1993m6 1995m2 1996m10 1998m6 2000m2 2001m10 2003m6 2005m2 2006m10 2008m6 2010m2 2011m10 2013m6

-2

Price of crude oil

CPI: Canada

CPI: US

Fig. 1. Relationship between the price of crude oil and the aggregate CPI.

5 4

Normalized data

3 2 1 0 -1

1961m10 1963m6 1965m2 1966m10 1968m6 1970m2 1971m10 1973m6 1975m2 1976m10 1978m6 1980m2 1981m10 1983m6 1985m2 1986m10 1988m6 1990m2 1991m10 1993m6 1995m2 1996m10 1998m6 2000m2 2001m10 2003m6 2005m2 2006m10 2008m6 2010m2 2011m10 2013m6

-2

Energy price index in the CPI basket (Canada) Price of crude oil

Energy price index in the CPI basket (US)

Fig. 2. Relationship between the price of crude oil and the CPI energy price index.

rates in lieu of normalised data (expressed in levels) in Fig. 3 does not change this conclusion as both series once again exhibit a significant degree of synchronised co-movements. A cursory look at the verticaldotted lines in 1974m1, 1986m2, 1990m8 and 2008m12 in Fig. 3 reveals specific large co-jumpings when the two economies were engulfed by major oil shocks or the 2007–2008 global financial crisis.2 Therefore,

2 This observation is also consistent with the view that consumers and firms are only responsive to large shocks as “the presence of costs to monitoring energy expenditures and of costs of adjusting consumption patterns might make households reluctant to respond to small energy price changes” Kilian (2008b, p.877).

while the relationship between the price of crude oil and the aggregate CPI has been weakened, the same thing cannot be said in relation to the existing nexus between the CPI-energy price sub-indices and oil prices. It is useful to examine the monthly mean percentage changes and standard deviation of the price of crude oil and various CPI sub-indices in the pre- and post-1999 periods in Figs. 4 and 5. It is apparent that on average monthly rises in the price of crude oil were 0.46% and 0.80% (almost double) per month during the pre- and post-1999 periods, respectively. Compared with other goods and services included in the CPI basket (particularly non-food and non-energy items), consumer energy prices for both countries show noticeably higher growth rates. In Canada the average monthly rises in the price of non-food

A. Valadkhani / Energy Economics 45 (2014) 33–44

37

Canada 100

1990m8

1986m2

1974m1

2008m12

75 50 25 0

10

-25

5

-50

%

0 -5 -10

1961m10 1963m6 1965m2 1966m10 1968m6 1970m2 1971m10 1973m6 1975m2 1976m10 1978m6 1980m2 1981m10 1983m6 1985m2 1986m10 1988m6 1990m2 1991m10 1993m6 1995m2 1996m10 1998m6 2000m2 2001m10 2003m6 2005m2 2006m10 2008m6 2010m2 2011m10 2013m6

-15

Price of crude oil (right axis)

Price of energy in the CPI basket (left axis)

US 1974m1

1986m2

100 2008m12

1990m8

75 50 25

20

0 10 -25

%

0

-50

-10 -20

1961m10 1963m6 1965m2 1966m10 1968m6 1970m2 1971m10 1973m6 1975m2 1976m10 1978m6 1980m2 1981m10 1983m6 1985m2 1986m10 1988m6 1990m2 1991m10 1993m6 1995m2 1996m10 1998m6 2000m2 2001m10 2003m6 2005m2 2006m10 2008m6 2010m2 2011m10 2013m6

-30

Price of crude oil (right axis)

Price of energy in the CPI basket (left axis)

Fig. 3. Percentage changes in the price of crude oil and the consumer energy price index.

0.800

0.80 0 0.70 0 0 0.60 0.50

0 0.50

0.46 0.444 0.3 38 0.35 0

0 0.40 0 0.30

0.38

0..35

0.34

0.223

0.38 0

0.38

0.37

0.21

0.220 0.17 0

0 0.20

0.17

3 0.13

0 0.10 0 0.00 Oil priices

Energy pricees (Caanada)

Eneergy prrices (US)

Food pprices (Canaada)

F Food pprices (US) (

19661m2-1999 9m122

N No nd on-foodd and Non-fo ood an non energy e y pricees no on enerrgy priices ( (Canad da) (U US)

CPII: Canaada

CPI: US

00m1--2013 3m6 200

Fig. 4. Mean percentage changes in the price of crude oil and various components of the CPI basket before and after 1999.

38

A. Valadkhani / Energy Economics 45 (2014) 33–44 8.57 7

9.000 8.000

6.99

7.000 00 6.0 5.000

3.992

00 4.0

3.25

3.000 1.550

00 2.0

1.49 9 1.13

0.76

0.7 77

1.000

00.41

0.40

00.19

0.43 3 0.39 9

N food aand Non-fo non energ e gy ppricees (US S)

CP PI: Caanadaa

0.28 8

0.3 30

00.41 0 0.35

0.0 00 Oil priices

gy priices Energ E (Caanadaa)

Energy pricees (US S)

Foood prices (Canada)

F Food pricees (U US)

1 1961m m2-1 1999m m12

Non n-foodd and d non n eneergy price p es (C Canadda)

CP PI: US S

2 2000m m1-22013m m6

Fig. 5. Standard deviations of percentage changes in the price of crude oil and various components of the CPI basket before and after 1999.

and non-energy items were 0.38% (pre-1999) and 0.13% (post-1999), while in the U.S. the corresponding growth rates were 0.38% versus 0.17%, respectively. However, the equivalent price rises in the energy component of the CPI were substantially higher in both pre- (Canada: 0.44% and the U.S.: 0.34%) and post-1999 (Canada: 0.35% and the U.S.: 0.50%) periods. Furthermore, according to Fig. 5 not only are energy prices in both Canada and the U.S. more volatile than other goods and services in the CPI basket (including food) but also the magnitude of such a volatility has noticeably risen over time. For example, while the pre-1999 standard deviation of energy prices in both Canada and the U.S. was almost the same (1.50% per month), the average post-1999 variation was more than double in Canada (3.25%) and the U.S. (3.92%). A quick glance at Figs. 4 and 5 reveals that although energy prices in Canada and the U.S. exhibit a similar upward trend, the extent of volatility and price rises are far greater in the U.S. than Canada.

4. Empirical results and main findings Table 1 presents the results of the four unit root tests [Augmented Dickey and Fuller (1981), Elliott et al. (1996), Phillips and Perron, 1988, and Kwiatkowski et al. (1992)] suggesting that both the price of crude oil and the CPI energy price indices are clearly I(1). The results of the Johansen cointegration test are shown in Table 2. Both the trace and maximal eigenvalue tests reject the null hypothesis of a zero cointegrating vector at the 1% level of significance. One can thus conclude that the CPI energy price index in each country and the price of crude oil are cointegrated. The DLS method is also employed to estimate Eq. (1) and obtain the resulting residual series or ECt from the corresponding cointegrating vector. The optimum number of leads and lags in Eq. (1) is determined using the Schwarz information criterion. Table 3 shows the results of the estimated short-run dynamic models for Canada and the U.S. by maximising the likelihood function in Eq. (12). The majority of the estimated key regime-switching coefficients (including the logistic coefficients appearing at the bottom of

Table 3) are statistically significant at the 5% level or better for each country and within each regime. In order to assess/choose the estimated regime switching models, this paper uses the following regime classification measure (RCM) of Ang and Bekaert (2002): RCM ðK ¼ 2Þ ¼ 400

T 1X p ð1−pt Þ T t¼1 t

ð13Þ

Where RCM denotes the regime classification measure, pi,t = p(st = i|ℑT) is the smoothed probability between the two regimes and K is the number of regimes. The constant term (i.e. 400) is used to normalize the RCM to lie between 0 and 100. The lower the RCM statistics, the better is the regime classification: when RCM = 0 it implies perfect regime classification and when RCM = 100 it means that no information about the regimes is revealed. Higher values of the RCM statistics indicate that the estimated models are subject to misspecification. Compared to the reported RCM values in Ang and Bekaert (2002), the estimated regime switching models in Table 3 have relatively low RCM statistics when K = 2 (Canada = 31.4 and the U.S. 19.64). q

As can be seen from the estimated∑ β0i in Regime 0, ceteris paribus, i¼0

the CPI energy price index will rise by 0.214% in Canada and 0.268% in the U.S. within the first 2 months if the price of crude oil increases by 1%. However, the equivalent short-run responses to a similar percentage change in the price of crude oil in Regime 1 are noticeably lower than q

Regime 0 as ∑ β1i is estimated to be 0.037% for Canada and 0.147% i¼0

for the U.S. Table 4 shows that for both countries the null hypothesis H 0 q

q

q

q

i¼0

i¼0

i¼0

i¼0

: ∑ β0i ¼ ∑ β1i vs. the alternative hypothesis H1 : ∑ β0i N∑ β1i is rejected at the 5% significance level or better. Therefore, one can argue that the positive effects of changes in the price of crude oil on energy

Table 1 Unit root test results. Test statistics

Ln(POt)

Xt = ΔLn(POt)

Canada (a)

ADF DF-GLS(b) PP(c) KPSS(d)

−2.364 −2.367 −2.158 0.307⁎⁎⁎

−19.720⁎⁎⁎ −19.660⁎⁎⁎ −19.268⁎⁎⁎ 0.056

Yt = ΔLn(Et)

Ln(Et)

−0.991 −1.064 −1.035 0.503⁎⁎⁎

US

Canada

US

−1.893 −1.866 −1.721 0.373⁎⁎⁎

−18.061⁎⁎⁎ −17.377⁎⁎⁎ −20.440⁎⁎⁎

−15.519⁎⁎⁎ −4.723⁎⁎⁎ −13.932⁎⁎⁎ 0.090

Note: (a) Augmented Dickey-Fuller (DF, 1981); (b) Elliott et al. (1996); (c) Phillips and Perron (1988); (d) Kwiatkowski et al. (1992). ⁎⁎⁎ Significant at the 1% level.

0.217

A. Valadkhani / Energy Economics 45 (2014) 33–44

39

Table 2 Johansen cointegration test results. Hypothesized no. of CE(s)

None At most 1

Canada

US

Eigenvalue

Trace statistic

Max. eigenvalue statistic

0.035 0.001

23.266⁎⁎⁎ 0.465

22.802⁎⁎⁎ 0.465

p-value(a) 0.00 0.50

Eigenvalue

Trace statistic

Max. eigenvalue statistic

p-value(a)

0.028 0.001

18.330⁎⁎⁎ 0.429

17.902⁎⁎⁎ 0.429

0.01 0.51

Note: (a) The reported p-values are based on MacKinnon et al. (1999). ⁎⁎⁎ Significant at the 1% level.

Table 3 Estimated short-run Markov-switching models. q

q

q

i¼0

i¼0

i¼1

EðY t jSt Þ ¼ ð1−St Þα 0 þ St α 1 þ ð1−St Þ∑ β0i X t−i þ St ∑ β 1i X t−i þ ð1−St Þ∑ γ0i Y t−i þ q

þSt ∑ γ1i Y t−i þ ð1−St Þλ0 EC t−1 þ St λ1 EC t−1 þ ut i¼1

Q t ¼ EðY t jSt Þ þ ð1−St Þσ 0 ut þ St σ 1 ut . Description

Canada Coefficient

Regime 0 α0 β00 β01 β02 λ0 γ01 Ln(σ0)

US z stat.

p-value

Coefficient

z stat.

p-value

0.001 0.187 0.081 −0.037 −0.147 0.383 −3.719

0.78 9.23 4.32 −1.66 −5.14 6.04 −71.25

0.44 0.00 0.00 0.10 0.00 0.00 0.00

0.001 0.045 0.070 0.032 −0.015 0.372 −4.951

2.34 6.87 6.89 4.36 −3.63 8.99 −92.48

0.02 0.00 0.00 0.00 0.00 0.00 0.00

2.248 −2.841 1.947 −5.719 19.64

7.05 −9.32

0.00 0.00

0.005 0.161 0.053

3.49 8.25 3.28

0.00 0.00 0.00

−0.035 0.125 −3.760

−3.30 2.35 −85.41

0.00 0.02 0.00

Regime 1 α1 β10 β11 β12 λ1 γ11 Ln(σ1)

0.001 −0.007 0.044

3.04 −1.59 5.09

0.00 0.11 0.00

−0.002 0.125 −5.413

−1.02 3.15 −62.71

0.31 0.00 0.00

Logistic intercept coefficients C00 C10 DW Schwarz criterion RCM

1.653 −1.391 2.024 −5.514 31.4

5.91 −5.95

0.00 0.00

prices in Regime 0 appear to be significantly larger and more instantaneous than Regime 1. The error correction coefficients have the correct sign as both λ0 and λ1 are negative. In Regime 0 the estimated feedback coefficients for Canada and the U.S. are −0.035 and −0.147, whereas in Regime 1 they are − 0.002 and − 0.015, respectively. Therefore, the error correcting speed of adjustment is much faster in Regime 0 than Regime 1, particularly for the U.S. In terms of dynamic inertia (i.e. the effect associated with the lagged dependent variable), there is not much difference between the two regimes and across the two countries. For Canada the coefficients of the first lagged-dependent variable in Regimes 0 and 1 are estimated to be the same (γ01 = γ11 = 0.125) and for the U.S. the difference between the corresponding coefficients is quite meagre (γ01 = 0.383 vs. γ11 = 0.372). Let us now examine the time varying-smoothed probability of switching from Regime 1 to Regime 0 in Fig. 6. It seems that for both countries the smoothed-probability of switching from Regime 1 to Regime 0 persistently remains at unity in the post-1999 period, whereas in the pre-1999 era such a switching appears to be non-systematic and less persistent, particularly in the context of the U.S. See the shaded versus non-shaded areas in Fig. 6. As discussed earlier, for both countries β00 N β01 + β02 and β10 b β11 + β12, one can thus conclude that in Regime 0 (mostly representing the post-1999 period) the positive effect of changes in the price of crude oil on energy prices is much higher and more instantaneous than Regime 1. Therefore, according to Tables 3 and 4 and Fig. 6, the marginal

effect of a given rise in the price of crude oil on the CPI energy prices has significantly and sustainably increased since 1999 for Canada and the U.S. To some extent this phenomenon can be attributed to the increasing weights of various energy components in the CPI. As can be seen from Table 5, for example the relative weights assigned to various components of the CPI (except utility gas service) in the U.S. have risen over time. This increase is particularly more noticeable in relation to the weights assigned to motor fuels which have more than doubled. Compared to the U.S., it appears that Canada has more irregular episodes of switching probabilities from the low marginal effect state (Regime 1) to the high marginal effect regime (Regime 0). This could partially be explained by (1) the fact that at the national level Canadian households consume more energy than their American counterparts; (2) Canada is a major oil-exporter whereas the U.S. is a major oilimporter3; (3) the composition of the sources of energy used is different in these two countries. For example, Canadian households use more electricity (40%) than the U.S. citizens (34%) for space and water heating (Maruejols et al., 2011). It should be noted that the year 2000 is an important turning point because in this year fossil fuel prices started rising substantially and concerns about the environmental consequences of greenhouse gas emissions also gathered momentum. The Western U.S. Energy Crisis 3 This could be a reason why the volatility and energy price rises in the U.S. are greater than those in Canada (see Figs. 4 and 5).

40

A. Valadkhani / Energy Economics 45 (2014) 33–44

Table 4 Testing for equality of the sum of marginal effects across the two regime. Country

q

q

i¼0

i¼0

H0 : ∑ β 0i ¼ ∑ β 1i

p-value

vs. q

q

i¼0

i¼0

H 1 : ∑ β 0i N∑ β 1i Canada U.S.

F(1,604) = 55.10⁎⁎⁎ F(1,604) = 5.57⁎⁎

0.00 0.02

⁎⁎⁎ Significant at the 1% level. ⁎⁎ Significant at the 5% level.

also occurred in 2000 whereby “prices for wholesale electricity tripled and then periodically spiked to unprecedented levels, even in December, with ample available capacity” (Hausman and Neufeld, 2011, p.740). It is interesting to note that the responsiveness of consumers and producers to energy price shocks has also undergone a significant

change since 2000. Kilian (2008b) simulated the responses of consumers and firms to the following three different types of retail energy price shocks on the U.S. economy: large versus small shocks; positive versus negative shocks; unprecedented versus predicated shocks. In three separate graphs Kilian (2008b, p.878) demonstrated that the dynamic responses to these price shocks consistently exhibited different behaviour in the post-1999 period. While the pre-1999 period showed random unsustained small fluctuations, the responses beginning in 2000 were consistently larger in magnitude and more oscillatory. One may also argue that the civil unrest in Venezuela in 2002, the Iraq war of 2003, and Hurricanes Katrina and Rita in 2005 also contributed to the switching probabilities remaining in regime 0. By causing many problems in energy distribution system and disrupting refineries, these events created a wedge between crude oil and energy prices. Can the above finding be substantiated using a different econometric approach? Table 6 presents the estimation results of the Bai–Perron sequential model using Eq. (3) allowing for one statistically significant break date. As can be seen from the results, the one statistically

Canada 1.2 1999m6-2013m2 Probability of switching to R0

1.0

0.8

0.6

0.4

0.2

1961m6 1963m2 1964m10 1966m6 1968m2 1969m10 1971m6 1973m2 1974m10 1976m6 1978m2 1979m10 1981m6 1983m2 1984m10 1986m6 1988m2 1989m10 1991m6 1993m2 1994m10 1996m6 1998m2 1999m10 2001m6 2003m2 2004m10 2006m6 2008m2 2009m10 2011m6 2013m2

0.0

US 1.2 1999m4-2013m2 Probability of switching to R0

1.0

0.8

0.6

0.4

0.2

1961m6 1963m2 1964m10 1966m6 1968m2 1969m10 1971m6 1973m2 1974m10 1976m6 1978m2 1979m10 1981m6 1983m2 1984m10 1986m6 1988m2 1989m10 1991m6 1993m2 1994m10 1996m6 1998m2 1999m10 2001m6 2003m2 2004m10 2006m6 2008m2 2009m10 2011m6 2013m2

0.0

Fig. 6. Smoothed probability of switching from regime 1 to regime 0 (using the WTI price measure of crude oil).

A. Valadkhani / Energy Economics 45 (2014) 33–44 Table 5 US city-average weights of major energy components in the CPI (%). Energy components

2011–2012

1993–1995

Gas (piped) and electricity Electricity Utility (piped) gas service Motor fuel Gasoline (all types) Other motor fuels All energy components All items less energy

4.22 3.34 0.89 6.61 6.48 0.13 11.07 88.93

3.88 2.71 1.17 3.19 3.17 0.02 7.31 92.69

Source: Bureau of Labor Statistics (2014).

significant and endogenously-estimated break date occurred exactly in the same year (June 1999 for Canada and April 1999 for the U.S.). The estimated pre- and post-1999 coefficients can loosely and pairwise be compared to the Markov-switching results in Regime 1 and 0, respectively. Despite the fundamental differences between these two approaches, the results of the Bai–Perron test also provide convincing evidence that the marginal impact of changes in the price of crude oil on consumer energy prices has significantly increased since 1999 for both Canada and the U.S. Ceteris paribus, a given rise in crude oil prices can now increase energy costs much faster than what was previously experienced. As another alternative approach, the time-varying marginal effects of changes in oil prices on the energy price index in Eq. (3) are also estimated using a rolling estimation window with 10 years of monthly data. Fig. 7 indicates that the estimated time-varying marginal effects   q i:e: ∑ βi for both countries exhibit an overall upward trend, partici¼0

ularly in the pre-2008 GFC era. The rise in the marginal effects demonstrated in Fig. 7 is consistent with the results of the regime switching model in Table 3 and the Bai–Perron approach (testing L + 1 vs. L sequentially determined break date) in Table 6. Despite the fundamental differences between these approaches, the results provide convincing evidence that the marginal effects of changes in the price of crude oil on consumer energy prices have significantly increased for both Canada and the U.S. since 1970. Based on the reported standard deviations in Fig. 5, energy prices in the U.S. and Canada exhibited almost the same volatility in the pre-2000

41

period (Canada = 1.50 and the U.S. = 1.49). However, while energy price volatilities rose for both countries in the post 2000, the magnitude of the rise was somewhat larger in the U.S. (3.92) than that in Canada (3.25). The above observations are quite consistent with the varying magnitudes of the regime-specific coefficients in Table 3 as in regime 0 both countries had similar standard error coefficients [Canada Ln(σ0) = −3.76 and the U.S. Ln(σ0) = −3.72]. However, in regime 1 (low marginal effects state) these coefficients were −5.41 and −4.95, respectively. These results thus justify the use of regime-specific error variances in the estimated Markov-switching models. It is important to check if the same results can be obtained if another alternative price measure for crude oil (other than WTI prices) is adopted. In order to examine the robustness of the final results, this paper thus uses the commodity PPI (Producer Price Index) for crude oil which is available in monthly frequency from the Bureau of Labor Statistics (http://data.bls.gov/cgi-bin/surveymost?wp). Comparing Fig. 6 (using the WTI measure of crude oil prices) with Fig. 8 (utilising the PPI measure of crude oil prices) reveals that the resulting smoothed probabilities of switching from regime 1 to regime 0 do not significantly change, particularly in the context of the U.S. In both Figs. 6 and 8 the smoothed probabilities persistently peak near unity in the post-1999 period and such a switching is not systematically sustained in the pre1999 period. Note how the probabilities sustainably rise in the shaded versus non-shaded areas for each country in Figs. 6 and 8. The consistency of the results in Figs. 6 and 8 is not surprising given the fact that the scatter and time plots of the normalised values for the above two price measures of crude oil (shown in Fig. 9) exhibit a very similar pattern with a substantial degree of synchronised co-movements throughout the sample period with the exception of the 2-year sub-period 1979– 1980. As with Figs. 1 and 2, both the PPI and WTI price measures in Fig. 9 are also shown using the corresponding normalised data so that we have the same measurement scale. In sum, the dynamic marginal effects of changes in the price of crude oil on consumer energy prices consistently increased and become more instantaneous for both countries in the post-1999 era compared to the pre-1999 period. These findings have direct policy implications for consumers as well as producers in the more energy-intensive sectors of the Canadian and U.S. economies. Since the estimated pass-through coefficients and the absolute values of the error correction coefficients for both countries have significantly risen since 1999, one can argue that

Table 6 q q Bai–Perron test results (L + 1 vs. L sequentially determined break dates). Y t ¼ α þ ∑ β i X t−i þ∑ γ i Y t−i þ λEC t−1 þ ut . i¼0

i¼1

Canada

US

Break: 1999 M06

Break: 1999 M04

Variable

Coefficient

t stat.

Sample α β0 β1 β2 λ γ1

1961 m07–1999 m05: 455 obs. 0.004 4.92 0.009 0.76 0.035 3.04

0.00 0.45 0.00

−0.016 0.057

0.00 0.29

Sample

1999 m06–2013 m02:165 obs.

α β0 β1 β2 λ γ1 R2 2 R DW Schwarz criterion

−0.003 0.217 0.136

−1.75 13.40 6.62

0.08 0.00 0.00

−0.131 0.085 0.360 0.351 1.969 −5.230

−5.85 1.56

0.00 0.12 0.00 0.02

−3.27 1.07

Note: Break selection is based on a trimming region of 0.15 and significance level of 0.05.

p-value

Coefficient

t stat.

1961 m07–1999 m03: 453 obs. 0.002 2.51 0.059 5.57 0.060 5.21 0.012 0.99 −0.024 −2.97 0.358 5.94

p-value 0.01 0.00 0.00 0.32 0.00 0.00

1999 m04–2013 m02: 167 obs. −0.005 0.215 0.134 −0.030 −0.212 0.352 0.591 0.583 1.940 −5.410

−3.35 14.78 7.55 −1.79 −8.55 7.65

0.00 0.00 0.00 0.07 0.00 0.00

42

A. Valadkhani / Energy Economics 45 (2014) 33–44

.5

Marginal effects

.4

.3

.2

.1

.0

1971m10 1973m6 1975m2 1976m10 1978m6 1980m2 1981m10 1983m6 1985m2 1986m10 1988m6 1990m2 1991m10 1993m6 1995m2 1996m10 1998m6 2000m2 2001m10 2003m6 2005m2 2006m10 2008m6 2010m2 2011m10 2013m6

-.1

CANADA

US

  q Fig. 7. Time-varying marginal effects i:e: ∑ βi of changes in oil prices on and the consumer energy price index. i¼0

changes in oil prices can now be passed onto consumers in larger magnitudes and with higher speed. Compared to the pre-1999 period, any rise in the price of crude oil in the post-1999 era can more quickly translate to higher fuel and transport costs as well as household utility bills (including electricity and natural gas). When oil prices exert more positive and immediate impacts on energy bills, monetary authorities will also have less reaction time to respond to inflationary pressures partially arising from unanticipated oil price shocks. As discussed earlier, if it was not for the adoption of more energy efficient technologies in various sectors of both economies, a sudden rise in crude oil prices could have increased energy costs much more than what it is observed in this study. 5. Conclusions Previous studies have highlighted the way in which oil prices influence inflation using a wide range of methodologies and data frequencies. Many economists believe that oil prices significantly contributed to CPI inflation in the U.S. particularly in the pre-1970 period. Both Canada and the U.S. have the highest per capita consumption of petroleum-based liquid fuel among all OECD countries (Knittel, 2012). The major objective of this paper is to analyse the dynamic relationship between the price of crude oil and the CPI energy-price sub-indices in these two countries employing a Markov-regime switching model (Hamilton, 1989, 1996) and the Bai and Perron (1998) sequential break-date detection approach. The use of different approaches ensures that our findings are robust and insensitive to the methodology adopted. According to the Johansen trace and maximum eigenvalue test results, both price series are cointegrated during the sample period, which spans from January 1961 to June 2013. Then, for each country a short-run dynamic model is estimated whereby all parameters (including the error-variance term) are allowed to switch across Regimes 0 and 1. The Bai–Perron approach is also employed to test L + 1 versus L sequentially determined breaks at the 5% level of significance. The empirical results from different approaches (including rolling regressions) consistently suggest that (a) the marginal effect of oil prices on the CPI energy-price sub-indices has shown a significant upward shift since 1999; and (b) the error-correction coefficient (the speed of adjustment) for both countries (particularly the U.S.) has also noticeably increased over the same period.

The smoothed probabilities are considered to quantify the likelihood of switching from Regime 1 (where the estimated marginal coefficients are relatively lower) to Regime 0 (where the estimated coefficients are relatively higher and more instantaneous). It is noteworthy that Regime 0 reflects the present situation for both countries because from 1999 onwards the probability of switching from Regime 1 to Regime 0 increased to near unity (see both Fig. 6 and Table 3). According to the Bai–Perron test results, the one statistically significant and endogenously estimated break date also occurred in the same year (June 1999 for Canada and April 1999 for the U.S.). Therefore, this paper provides compelling evidence that the marginal impact of changes in oil prices on consumer energy prices has not only exhibited a sustained rise since 1999 but also such dynamic influences have now become more instantaneous. Put differently, the results reveal that the estimated coefficients associated with the marginal effects of the lagged oil price changes on energy prices have become much smaller than the impacts arising from instantaneous changes in oil prices. Ceteris paribus, a given rise in crude oil prices can now increase consumer energy costs much faster than what was previously experienced. Compared with other goods and services included in the CPI basket (particularly non-food and non-energy items) oil prices exhibited noticeably higher growth rates and volatilities in the post- rather than pre-1999 periods. Bachmeier and Cha (2011, pp. 1166–67) argue that “the large oil price increases of the 1970s led to the adoption of energy-saving technologies, reducing the importance of oil as a factor of production relative to capital, labor, and materials”. Therefore, if it was not for such energy-saving technologies, the rising marginal effects of oil prices on consumer energy costs (i.e. fuel, transport as well as utilities including electricity and natural gas) could have raised the aggregate CPI more than what it is currently observed. The above significant jump in the marginal effects of oil prices on energy costs can be somewhat explained by the global awareness about the environmental consequences of greenhouse gas emissions, as well as the rising fossil fuel prices and the Western U.S. Energy Crisis of 2000.

Appendix A. Supplementary data Supplementary data to this article can be found online at http://dx. doi.org/10.1016/j.eneco.2014.06.015.

A. Valadkhani / Energy Economics 45 (2014) 33–44

43

Canada 1.2 1999m6-2013m2

Probability of switching to R1

1.0

0.8

0.6

0.4

0.2

1963m2 1964m10 1966m6 1968m2 1969m10 1971m6 1973m2 1974m10 1976m6 1978m2 1979m10 1981m6 1983m2 1984m10 1986m6 1988m2 1989m10 1991m6 1993m2 1994m10 1996m6 1998m2 1999m10 2001m6 2003m2 2004m10 2006m6 2008m2 2009m10 2011m6 2013m2

0.0

US 1.2 1999m4-2013m2

Probability of switching to R1

1.0

0.8

0.6

0.4

0.2

1963m2 1964m10 1966m6 1968m2 1969m10 1971m6 1973m2 1974m10 1976m6 1978m2 1979m10 1981m6 1983m2 1984m10 1986m6 1988m2 1989m10 1991m6 1993m2 1994m10 1996m6 1998m2 1999m10 2001m6 2003m2 2004m10 2006m6 2008m2 2009m10 2011m6 2013m2

0.0

Fig. 8. Smoothed probability of switching from regime 1 to regime 0 (using the PPI price measure of crude oil).

References Ajmera, R., Kook, N., Crilley, J., 2012. Impact of commodity price movements on CPI inflation. Mon. Labour Rev. 29, 29–43. Ali Ahmed, H.J., Wadud, I.K.M.M., 2011. Role of oil price shocks on macroeconomic activities: an SVAR approach to the Malaysian economy and monetary responses. Energy Policy 39 (12), 8062–8069. Ang, A., Bekaert, G., 2002. Regime switches in interest rates. J. Bus. Econ. Stat. 20, 163–182. Bachmeier, L.J., Cha, I., 2011. Why don't oil shocks cause inflation? Evidence from disaggregate inflation data. J. Money Credit Bank. 43 (6), 1165–1183. Bai, J., 1997. Estimating multiple breaks one at a time. Econ. Theory 13, 315–352. Bai, J., Perron, P., 1998. Estimating and testing linear models with multiple structural changes. Econometrica 66, 47–78. Broyden, C.G., 1970. The convergence of a class of double-rank minimization algorithms. J. Inst. Math. Appl. 6, 76–90. Burbridge, J., Harrison, A., 1984. Testing for the effects of oil-price rises using vector autoregressions. Int. Econ. Rev. 25 (1), 459–484. Bureau of Labor Statistics, 2014. Relative importance of components in the Consumer Price Indexes: U.S. city average. http://www.bls.gov/cpi/cpiri_2013.pdf (accessed 20 March 2014). Cologni, A., Manera, M., 2008. Oil prices, inflation and interest rates in a structural cointegrated VAR model for the G-7 countries. Energy Econ. 30, 856–888.

De Gregorio, J., Landerretche, O., Neilson, C., 2007. Another pass through bites the dust? Oil prices and inflation. Economia 7, 45–65. Dickey, D., Fuller, W., 1981. Likelihood ratio tests for autoregressive time series with a unit root. Econometrica 49, 1057–1072. Econdata, 2013. DX Data: IFS Database. EconData Pty Ltd, Melbourne. Elliott, G., Rothenberg, T., Stock, J., 1996. Efficient tests for an autoregressive unit root. Econometrica 64 (4), 813–836. Filis, G., 2010. Macro economy, stock market and oil prices: do meaningful relationships exist among their cyclical fluctuations? Energy Econ. 32 (4), 877–886. Fletcher, R., 1987. Practical Methods of Optimization, 2nd ed. John Wiley & Sons, New York. Goldberg, P.K., 1998. The effects of the corporate average fuel efficiency standards in the U.S. J. Ind. Econ. 46 (1), 1–33. Hamilton, J.D., 1983. Oil and the macroeconomy since World War II. J. Polit. Econ. 92 (2), 228–248. Hamilton, J.D., 1989. A new approach to the economic analysis of nonstationary time series and the business cycle. Econometrica 57, 357–384. Hamilton, J.D., 1996. Specification testing in Markov-switching time-series models. J. Econ. 70, 127–157. Hausman, W.J., Neufeld, J.L., 2011. How politics, economics, and institutions shaped electric utility regulation in the United States: 1879–2009. Bus. Hist. 53 (5), 723–746. Hayashi, F., 2000. Econometrics. Princeton University Press, Princeton.

44

A. Valadkhani / Energy Economics 45 (2014) 33–44

5

4

Normalised data

3

2

1

0

1961m6 1963m2 1964m10 1966m6 1968m2 1969m10 1971m6 1973m2 1974m10 1976m6 1978m2 1979m10 1981m6 1983m2 1984m10 1986m6 1988m2 1989m10 1991m6 1993m2 1994m10 1996m6 1998m2 1999m10 2001m6 2003m2 2004m10 2006m6 2008m2 2009m10 2011m6 2013m2

-1

Price of oil (WTI) Price of oil (PPI from BLS)

Price of oil (PPI from BLS): Normalised data

5

4

3

2

1

0

-1 -1

0

1

2

3

4

5

Price of oil (WTI): Normalised data

Fig. 9. Normalised time and scatter plots of the PPI and WTI price measures of crude oil.

Hooker, M.A., 2002. Are oil shocks inflationary? Asymmetric and nonlinear specifications versus changes in regime. J. Money Credit Bank. 34, 540–561. Katayama, M., 2013. Declining effects of oil price shocks. J. Money Credit Bank. 45 (6), 977–1016. Kilian, L., 2008a. A comparison of the effects of exogenous oil supply shocks on output and inflation in the G7 countries. J. Eur. Econ. Assoc. 6 (1), 78–121. Kilian, L., 2008b. The economic effects of energy price shocks. J. Econ. Lit. 46 (4), 871–909. Kim, C.J., 1994. Dynamic linear models with Markov-switching. J. Econ. 60, 1–22. Knittel, C.R., 2012. Reducing petroleum consumption from transportation. J. Econ. Perspect. 26 (1), 93–118. Kwiatkowski, D., Phillips, P.C.B., Schmidt, P., Shin, Y., 1992. Testing the null hypothesis of stationarity against the alternative of a unit root. J. Econ. 54, 159–178. MacKinnon, J.G., Haug, A.A., Michelis, L., 1999. Numerical distribution functions of likelihood ratio tests for cointegration. J. Appl. Econ. 14, 563–577. Maruejols, L., Lu, X., Young, D., 2011. A Comparison of Energy-Related Characteristics of Residential Dwellings and Technologies across Canada and the US, Canadian Building Energy End-Use Data and Analysis Centre, CBEEDAC 2011-RP-05, Alberta Canada. Miller, J.I., Ratti, R.A., 2009. Crude oil and stock markets: stability, instability, and bubbles. Energy Econ. 31 (4), 559–568.

Nixon, D., Smith, T., 2012. What can the oil futures curve tell us about the outlook for oil prices? Q. Bull. 52 (1), 39–47. Phillips, P.C.B., Perron, P., 1988. Testing for a unit root in time series regression. Biometrika 75, 335–346. Ratti, R.A., Hasan, M.Z., 2013. Oil price shocks and volatility in Australian stock returns. Econ. Rec. 89, 67–83. Ratti, R.A., Vespignani, J.L., 2013. Crude oil prices and liquidity, the BRIC and G3 countries. Energy Econ. 39, 28–38. Regnier, E., 2007. Oil and energy price volatility. Energy Econ. 29 (3), 405–427. Stock, J., Watson, M., 1993. A simple estimator of cointegrating vectors in higher order integrated systems. Econometrica 61 (4), 783–820. Valadkhani, A., 2014. Switching impacts of the output gap on inflation: evidence from Canada, the UK and the US. Int. Rev.f Econ. Financ. 33, 270–285. Venditti, F., 2013. From oil to consumer energy prices: how much asymmetry along the way? Energy Econ. 40, 468–473. Wang, Y.S., 2013. Oil price effects on personal consumption expenditures. Energy Econ. 36, 198–204.