A Structural VAR-Panel ARDL Approach to Oil Price Shocks and Oil Import Demand

Transcription

1 A Structural VAR-Panel ARDL Approach to Oil Price Shocks and Oil Import Demand Mikidadu Mohammed Department of Economics The University of Utah, Salt Lake City, UT Abstract Are all oil price shocks the same? How sensitive are oil-importing countries to oil price shocks? As it stands, the mechanisms underlying the transmission of oil price shocks to oil-importing countries are not adequately understood. Part of the reason lies in the lack of understanding of the causes of oil price shocks. Indeed, without knowing what drove up the price of oil in the first place, it becomes almost impossible to accurately predict its implications. Combining the structural VAR model proposed by Kilian and the panel ARDL model by Pesaran, Shin, & Smith, this study develops a two-step theoretical and empirical framework to determine the underlying shocks to the real price of oil and estimate the effects of the shocks on oil import demand, the key macroeconomic variable responsible for the transmission of oil price shocks to oil-importing countries. Relying on more recent data ( ), I provide further evidence that not all oil price shocks are the same; they originate from different sources and thus have different consequences. In addition, I refine the traditional modeling and interpretation of the effects of oil price shocks on oil import demand. More specifically, I model import demand for oil crude and petroleum products not as a function of a single oil price shock as standard in existing oil import demand literature but as a function of three distinct oil price shocks resulting from oil supply shocks, aggregate demand shocks, and precautionary demand shocks. For the 31 oil-importing countries included in the sample, I find that import demand for oil is much less sensitive to oil price shocks driven by oil supply shocks but highly sensitive to oil price shocks driven by aggregate demand and precautionary demand shocks. However, while rich and poor oil-importing countries are almost identical in their response to oil supply shocks, they differed remarkably in their response to aggregate demand and precautionary demand shocks. Keywords: oil price shocks, structural VAR, panel ARDL, import, demand, elasticity JEL Classification: C1, C5, C33, C51, Q4, Q41, Q43 1 Introduction The literature on import demand for oil is replete with studies that examine the demand for oil in general and raw crude or petroleum products in particular. Atkins & Jazayeri (2004) provide an excellent survey on earlier studies. Recent studies conducted in a time-series context include Cooper (2003) on US and 22 other countries; Altinay (2007) on Turkey; Gosh (2009) on India; Ziramba (2010) on South Africa; Sun et al. (2010) on China; Asali (2011) on G7 and BRIC countries; and Kim & Baek (2013) on South Korea. Recent studies conducted in pooled cross-section or panel context include Dargay & Gately (2010) for a pooled cross-section of six groups of countries; IMF (2011) for a panel of 45 OECD and non-oecd countries; and Javan & Zahran 1

2 (2015) for a panel of 25 OECD and non-oecd countries that represent 75% of global oil demand. The elasticity estimates from these country-specific and panel studies vary substantially. Nonetheless, the overall empirical results point to one direction: oil import demand is generally price and income inelastic in the short run (Table 1). A standard approach utilized by these studies is to model oil import demand as a function of income (GDP) and the price of oil. One reason for this modeling approach is that, fluctuations in the real price of oil is thought to originate from oil supply disruptions driven by events in the Middle East. Since these events are outside the confines of macroeconomic models, they are, together with oil price fluctuations, treated as exogenous. Another reason relates to the notion that, oil import demand for any single country is not large enough to affect world oil price; hence the price of oil is assumed to be exogenously determined. Recent research on oil shocks have challenged these long-held beliefs about the causes of oil price shocks and provided strong theoretical and empirical evidence that, oil supply disruptions is not the only driver of fluctuations in the real price of oil, but other factors such as dynamics in global macroeconomic activity as well as concerns about future oil supply shortfalls, also influence the real price of oil (Kilan 2009a,b; Kilan & Murphy, 2012; Alquist et al., 2013; Kilan, 2014). Indeed, since Kilian (2009a), it has become evident that oil price shocks are driven by distinct demand and supply shocks: shocks from global crude oil production caused by oil supply disruptions associated with events in oil-producing countries (oil supply shocks); shocks to the global demand for all industrial commodities including oil (aggregate demand shocks); and shocks driven by shifts in precautionary demand for oil due to uncertainties about the availability of future oil supply (precautionary demand shocks). Consequently, current research efforts have been directed towards the reexamination of the effects of these distinct oil price shocks on macroeconomic variables such as output and inflation, but none on oil import demand. Thus, the aim of this paper is to extend the recent advances in the oil shocks literature to develop a theoretical and empirical framework to examine the effects of the three distinct oil price shocks on import demand for raw crude oil and petroleum products. The analysis relies on more recent data ( ) for a panel of 31 oilimporting countries. The central message of the paper is that the standard approach to modeling oil import demand as a function of oil price and GDP is not well-defined because it implicitly assumes that one can vary the price of oil while holding all other variables fixed. This ceteris paribus assumption is not appropriate for two reasons. First, the standard approach implies that oil import demand is endogenous while the price of oil and GDP are exogenous. But this is not the case since the price of oil and GDP are potentially endogenous and reverse causality may run from one to the other. Second, shocks to the global demand for all industrial commodities may have direct effects on oil import demand as well as an indirect effect working through the price of oil. Thus, even if one controls for the reverse causality, these direct and indirect effects of global demand shocks on oil import demand invalidate the ceteris paribus assumption. The paper jointly addresses these issues using the structural VAR model proposed 2

3 by Kilian (2009a). In particular, it identifies the underlying demand and supply shocks in the global crude oil market and decomposes the real price of oil into three components: oil supply shocks, aggregate demand shocks, and precautionary demand shocks. Consistent with the oil shocks literature, oil supply shocks are found to have historically made relatively small contributions to fluctuations in the real price of oil while the largest contributions have come from aggregate demand shocks and precautionary demand shocks. The identification of these shocks is crucial not just for explaining fluctuations in the real price of oil but also for understanding the degree of responsiveness elasticities of oil import demand associated with oil price fluctuations. Relying on the panel autoregressive distributed lag (ARDL) approach to cointegration, the study estimates the price elasticities of import demand for crude and petroleum products to each of the decomposed oil price shock. At the panel level, the short run price elasticity estimates reveal that an adverse oil supply shock causes a mild transitory decrease in import demand for crude and petroleum products while an aggregate demand expansion causes a delayed but strong persistent increase in import demand for crude and petroleum products. Precautionary demand shock triggers an immediate increase in petroleum products import more than raw crude import. The country level elasticity estimates generally reflect the panel estimates. However, there are subtle variations within and across rich and poor oil-importing countries. The analysis presented in this paper differs from those in existing oil import demand literature in two key respects. First, it endogenizes the price of oil and decomposes oil price shocks into three distinct components. Second, it models crude and petroleum import demand as a function of the decomposed oil price shocks and provides the short run price elasticity estimates to each shock as well as the time path of the elasticities. In these respects, the analysis shows that, contrary to the standard modeling approach in existing oil import demand literature, crude and petroleum import demand do not depend on a single oil price shock but depends on multiple oil price shocks resulting from oil supply shocks, aggregate demand shocks, and precautionary demand shocks. The rest of the paper is organized as follows. Section 2 lays out the methodology utilized to determine the underlying demand and supply shocks to the real price of oil and examine their respective impact on import demand for crude and petroleum products. Section 3 describes the data. Results are presented and discussed in Section 4. Section 5 concludes. 2 Methodology The paper develops a two-step theoretical and empirical framework to address the short run effects of oil price shocks on import demand for crude and petroleum products. First, the structural VAR model proposed by Kilian (2009a) is utilized to determine the underlying demand and supply shocks to the real price of oil. Second, the degree of responsiveness elasticities of crude and petroleum import demand to the underlying demand and supply shocks from the first step are examined using the Pesaran, Shin, & Smith (1999) panel ARDL model. 3

4 2.1 The Structural VAR model Following Kilian (2009a), the study defines a structural VAR model of the global crude oil market as: 24 A 0 z t = α + A i z t 1 + ε t (1) i=1 where z t denotes a vector time series consisting of the percent change in monthly global crude oil production ( P ROD t ), a detrended index of monthly global real economic activity in industrial commodity markets (REA t ), and the monthly real oil price (ROP t ). REA t and ROP t are expressed in logs. The model allows for two years of lags. The sample period covers 1980:1 to 2013:12. Figure 1 plots P ROD, Figure 2 plots REA, and Figure 3 plots ROP. The error term, ε t, denotes the vector of serially and mutually uncorrelated structural innovations. A 1 0 has a recursive structure such that the reduced-form errors e t can be decomposed according to e t = A 1 0 ε t : e t e prod t e rea t e rop t a = a 21 a 22 0 a 31 a 32 a 33 where ε 1t denotes shocks to the global supply of crude oil (oil supply shock); ε 2t denotes demand shocks to the global demand for all industrial commodities including oil (aggregate demand shock); and ε 3t denotes demand shocks specific to the global crude oil market driven by precautionary demand for oil (precautionary demand shock). The structural model postulates that the real price of oil is determined by the intersection of a vertical supply curve of oil and a downward sloping demand curve for oil. Oil demand shocks (driven by aggregate demand shock or precautionary demand shock) move the demand curve along the vertical supply curve causing the real price of oil to change. Oil supply shocks triggered by say, a disruption in global oil supply move the vertical supply curve along the demand curve, again causing the real price of oil to change. As in Kilian (2009a,b), the following restrictions were imposed on A 1 0. First, it is assumed that oil supply will not respond to innovations to the demand for oil within the same month. This exclusion is conceivable as in practice, oil-producing countries tend to be slow in their response to demand shocks due to increased adjustment costs of oil production and the uncertainty about the state of the crude oil market. Second, increases in the real price of oil driven shocks that are specific to the oil martket will not lower global real economic activity immediately, but with a delay of at least one month, as it takes time for global real economic activity to adjust. Finally, by construction, shocks to the real price of oil that cannot be explained by oil supply shocks or aggregate demand shocks are attributed to changes in the precautionary demand for oil due to uncertainties about the availability of future oil supplies. The structural innovations (shocks) implied by the VAR model are in monthly frequency. And since the main concern of the study is to examine how each of the decomposed oil price shocks affect crude and petroleum import (data on which is available only on annual basis for majority of the countries in the sample), the study ε 1t ε 2t ε 3t 4

5 constructs annual shock series by averaging the monthly oil price shocks for each year as follows: ˆζ s,t = i=1 ˆε s,i,t, s = 1, 2, 3 where ˆζ s,t, s = 1,2,3 are the three decomposed oil price shocks and ˆε s,i,t refers to the estimated residual for the s th oil price shock in the i th month of the t th year of the sample. Figure 4 depicts time path of the annualized oil price shocks implied by the model. As indicated by the plots, at any giving point in time, fluctuations in the real price of oil is driven by a multitude of shocks: shocks from the global crude oil production (oil supply shock), shocks to the global demand for all industrial commodities including oil (aggregate demand), and shocks driven by shifts in the precautionary demand for oil (precautionary demand shocks). Having carefully delineate the underlying demand and supply shocks to the real price of oil, the study proceeds to examine how each of the decomposed oil price shock affect import demand for crude and petroleum products. The study considers three response variables: raw crude import bill as share of GDP (henceforth, crude import demand), petroleum products import bill as share of GDP (henceforth, petroleum import demand), and crude plus petroleum (oil) import bill as share of GDP (henceforth, oil import demand). Detailed descriptions of the variables are provided in Section 3. One approach to examining the effects of the three decomposed oil price shocks on the three response variables is the use of a distributed lag regression model of the following form: M t = α s + ψ s,i 5 ˆζ s,t i + ɛ s,t, s = 1, 2, 3 (2) i=1 where M t, denotes the respective fuel import demand; ψ s,i are the impulse response coefficients which corresponds to horizon i; ˆζ s,t i, s = 1,2,3 are the lags of the three serially uncorrelated oil price shocks; and ɛ s,t are potentially serially correlated errors. Kilian (2009b) utilized this distributed lag regression model on an equation-byequation basis to examine the effects of oil shocks on external balances for individual and groups of countries. Though dynamic, the distributed lag equation-by-equation approach is less parsimonious especially if the study is dealing with a large set of countries. In addition the equation-by-equation approach does not fully incorporate the sample dynamics since each country or group is modelled separately. An alternative modeling technique adopted by this paper is the panel ARDL approach to cointegration. Unlike the equation-by-equation model, the panel ARDL model takes the sample dynamics into account by using the lags of the dependent variable and the lagged and contemporaneous values of the explanatory variable(s) to estimate a single equation that concurrently provides elasticity estimates at the panel level as well as the country-specific level. An added advantage of the panel ARDL model is that, it can be applied irrespective of whether the underlying variables in the model are purely I (0) or purely I (1) or partially integrated (Pesaran, Shin, & Smith 1999; Altinay, 2007; von Arnim & Prabheesh, 2013). 5

6 2.2 The Panel ARDL approach to cointegration The panel ARDL approach to cointegration was developed by Pesaran, Shin, & Smith (1999). To estimate the elasticities of crude, petroleum products, and oil import demand to the decomposed structural shocks, the following panel ARDL(p,q) model is constructed: M i,t = p λ i,j M i,t j + j=1 q ψ i,j ˆζs,t j + µ i + ɛ i,t p = 1; q = 5 (3) j=0 where M i,t denotes the respective fuel import demand; ˆζs,t, s = 1,2,3 are the three decomposed oil price shocks; µ i is the fixed effects; ɛ i,t is the error term and are independently distributed across i and t with means 0 and variance σi 2 > 0; the coefficients of the lagged dependent variable, λ i,j, are scalars; ψ i,j are k x 1 coefficient vectors of the oil price shocks; i = 1, 2,.., N ; and t = 1, 2,.., T. Reparameterizing (3) yields the following error correction model to be estimated: p 1 q 1 M i,t = φ i (M it 1 β i ˆζs,t 1 )+ λ i,j M i,t j + ψ i,j ˆζ s,t j +µ i +ɛ i,t p = 1; q = 5 j=1 (4) where φ i is the error-correction speed of adjustment parameter; (M i,t 1 - β i ˆζs,t 1 ) represents the long run component of the model; and ψ i,j are k x 1 vector of short run elasticities of the response variables to the three decomposed oil price shocks. The parameter q is chosen to coincide with the maximum horizon of the short run price elasticities to be estimated. As standard in the literature for short run analysis, the maximum horizon is set to five years. To capture the time path of the short run response to the decomposed oil price shocks, the short run price elasticities are not summed across the five years worth of lags. Rather they are presented for each period as estimated by the model. Also, a time trend is included in (4) to account for trends in each of the response variables overtime. Of primary interest are the error-correction speed of adjustment parameter, φ i, the long run oil price elasticity coefficients, β i, and the short run oil price elasticity coefficients, ψ i,j. Theoretically, φ i is expected to be negative if the variables exhibit a return to long-run equilibrium. For an adverse oil supply shock, the long run price elasticities of import demand, β i, and short run price elasticities of import demand, ψ i,j, are expected to be negative and the absolute value < 1 (inelastic). For demand shocks (positive aggregate demand shock and adverse adverse precautionary demand shock), the long run price elasticities of import demand, β i, and short run price elasticities, ψ i,j, are expected to be positive. j=0 3 Data Table 2 provides summary data definitions and sources. Summary statistics of all the variables are reported in Table 3. The series used to generate the structural shocks are 6

7 monthly global crude oil production (PROD), a detrented monthly index of real economic activity (REA), and the monthly real oil price (ROP). PROD data was collected from the US Energy Information Administration (EIA); REA is derived based on Kilian (2009a) with current data from Kilian s research webpage; and ROP is computed based on refiner acquisition cost of imported crude oil from the US EIA and deflated by the US CPI. The sample period for the monthly series is 1980:1 2013:12. To generate the structural shocks, each of the monthly series was transformed into the following: percent change in monthly global crude oil production ( PROD), log of monthly real economic activity index (LG-REA), and log of monthly real oil price (LG-ROP). The Augmented Dickey-Fuller test indicates that all the monthly series in their transformed versions are stationary except real oil price (Table 4). The structural innovations (shocks) implied by the VAR model are in monthly frequency. Since the main concern of the study is to examine how each of the oil price shocks affect crude and petroleum import (data on which is available only on annual basis for majority of the countries in the sample), annualized shock series were constructed by averaging the monthly oil price shocks for each year as discussed in Section 2.1. Regarding the response variables, annual crude and petroleum import data, in million tons of oil equivalent (mtoe), was collected from the International Energy Agency (IEA) for a sample of 31 oil importing countries from A country is classified as an oil importer if its average volume of oil exports is less than 30% of its average volume of oil imports during Oil importing countries included in the study are: United States, Austria, Bangladesh, Belgium, Morocco, Chile, Portugal, Finland, France, Germany, India, Ireland, Israel, Italy, Japan, Jordan, Kenya, South Korea, the Netherlands, New Zealand, Pakistan, Dominican Republic, El Salvador, Switzerland, Philippines, South Africa, Spain, Sri Lanka, Sweden, Turkey, and Thailand. The import data is converted from mtoe to million barrels of oil per year (mbpy) based on Brent crude conversion factor of 1 ton of oil equivalent equals 7.57 barrels of oil equivalent. Crude, petroleum, and oil import, in mbpy, were each multiplied by the nominal annual crude oil price and each expressed as a share of GDP to determine the three response variables: crude import demand, petroleum import demand, and crude plus petroleum (oil) import demand. Since the response variables are for a panel of countries, a panel unit root test was applied to examine stationarity. However, before applying the panel unit root test, an important step is to first determine whether the sample data is cross-sectional dependent or independent. The standard test for cross-sectional dependence is the Pesaran (2004) CD test. Testing cross-sectional dependence is crucial because its presence implies that there is a common structure among observations in the sample that needs to be accounted for. Also, ascertaining cross-sectional dependence or independence is important since appropriate panel unit root tests have been developed for each case. For cross-sectional independence in panels, the appropriate panel unit root tests are Levin, Lin, & Chu (2002) LLC test and Phillips & Perron (1988) PP test. However, under cross-sectional dependence, there is a tendency for LLC and PP tests to lead to the over-rejection of the null of cross-sectional independence. Thus, in cases where cross-sectional dependence 7

8 is detected, the Pesarans (2007) CIPS test is appropriate as it uses a common structure to account for cross-sectional dependence (Chang, 2015). Table 5 reports the results from the Pesaran (2004) CD test on the respective fuel import demand. The results indicate that the null hypothesis of cross-sectional independence can be rejected at the 5% level of significance. Thus, the study applied the Peseran (2007) CIPs test, the appropriate panel unit root for a cross-sectional dependent sample. Results in Table 6 indicate that data on the respective fuel import demand are all stationary. 4 Results 4.1 Empirical results from the structural VAR model How do the real price of oil respond to demand and supply shocks in the global crude oil market? Figure 5 depicts the response of the real price of oil to the three structural shocks ˆε s,t, s = 1, 2, 3. The plots show that fluctuations in the real price of oil originate from different sources. Consistent with the findings in Kilian (2009a,b), an unexpected oil supply disruption causes a small transitory increase in the real price of oil within the first year (panel 1); an unanticipated increase in aggregate demand for all industrial commodities causes a delayed but persistent increase in the real price of oil (panel 2); and an unanticipated increase in precautionary demand for oil triggers an immediate, large, and persistent increase in the real price of oil (panel 3). As in Alquist & Kilian (2010), the real price of oil is found to overshoot in response to precautionary demand shocks Cumulative effect of oil demand and oil supply shocks on the real price of oil Figure 6 shows the historical decompositions of the real price of oil based on model (1). It is evident from the first panel that oil supply shocks have historically made relatively small contributions to fluctuations in the real price of oil. Indeed, the largest contributions come from aggregate demand shocks and precautionary demand shocks (second and third panels). While aggregate demand shock is responsible for longer swings in the real price of oil, precautionary demand shock is responsible for relatively sharp swings in the real price of oil. These results are consistent with the findings in the oil shocks literature (Barsky & Kilian, 2004; Kilian 2009a,b; Knittel & Pindyck, 2013; Fattouh et al., 2013; Kilian & Murphy, 2014; Kilian, 2014). Adding to the oil shocks literature, our decomposition shows that the increase in the real price of oil between 2009 and 2010, is driven in most part by aggregate demand shock despite the sluggish global real economic activity post-2008 recession. However, from 2011 to 2013, the rise in the real price of oil is driven mainly by precautionary demand for oil due to uncertainties about future oil supply shortfalls. 8

9 Indeed, our estimates from the structural VAR model provide further evidence that not all price shocks are the same; they originate from different sources and thus are likely to have different consequences. 4.2 Empirical results from the panel ARDL model Cointegration test results The ARDL approach to cointegration requires that the underlying variables in the model are cointegrated. This requirement ensures that the model reaches a long-run stable equilibrium. To examine cointergration, the study applied three panel cointegration tests (Fisher, Pedroni, and Kao). Results in Table 7 indicate that the annualized oil shocks series and the respective fuel import demand for all models to be estimated are cointegrated based on at least two of three panel cointegration tests. Having established the cointegration relationship among the annualized oil price shocks series and the respective fuel import demand, the long run and short run elasticities of the ARDL model in (4) were estimated Panel results Tables 8, 9, and 10 report the panel results for each response variable to each of the decomposed oil price shocks. The long run price elasticities of crude, petroleum, and oil import demand are reported in the first part of each table while the short run dynamics are reported in the second part. The elasticities represent the degree of responsiveness to a 10% structural innovation in each of the decomposed oil price shocks. As expected, the error-correction speed of adjustment parameter, φ i, is negative and significant in all estimated models, confirming that the underlying variables in the model exhibit a return to long-run equilibrium. Also, as expected, the panel long run price elasticities of the respective fuel import demand, β i, are negative and inelastic in response to an adverse oil supply shock although not statistically significant (Table 8). For a positive aggregate demand shock and precautionary demand shock, the long run price elasticities of import demand have the theoretically unexpected signs and are not statistically significant (Tables 9 and 10). However, crude import demand s response to precautionary demand shock has the theoretically expected sign and is statistically significant at the 10% level (Table 10 Model 7). Figure 7 plots the panel short run elasticities of each response variable to the three decomposed oil shocks from Tables 8,9, and 10. All the short run elasticity estimates (with few exceptions in Models 1, 2, 5, 6 and 7) are statistically significant at the 5% level. For a 10% structural innovation in each of the three distinct oil price shocks and a maximum elasticities horizon of five years, the panel elasticity estimates reveal the following results. In response to an adverse oil supply shock, the price elasticity of import demand for crude and petroleum products is negative and price inelastic. This suggests that, an adverse oil supply shock results in a less than proportionate decrease in import demand for crude and petroleum products on impact but the decrease in import demand is transitory as the elasticity turns positive around the second year. 9

10 In response to a positive aggregate demand shock, the price elasticity of import demand for crude and petroleum products is persistently positive and price inelastic implying that, a positive aggregate demand shock results in a relatively strong increase in import demand for crude and petroleum products with a year delay. The increase in imports peaks about the third year and begins to decline afterwards. Precautionary demand for oil due to uncertainties about the availability of future oil supply triggers an immediate increase in petroleum products import more than raw crude import. Both petroleum products and crude import peak in about a year and decline subsequently as concerns about future oil supply shortfalls ease. The finding that precautionary demand for oil triggers an immediate increase in petroleum products import more than raw crude import is suggestive and appears undetected in the oil import demand literature Country-level results Tables 11 to 19 summarize the country-level short run results. All the short run price elasticity estimates are statistically significant at the 5% level. Figures 8, 9, and 10 plot the short run elasticities for three rich oil-importing economies (US, Germany, and the Netherlands); Figures 11, 12, and 13 plot the short run elasticities for three middle-income oil-importing economies (South Korea, Thailand, and South Africa); and Figures 14, 15, and 16 plot the short run elasticities for three poor oil-importing economies (Bangladesh, Kenya, and El Salvador). 1 Plots for the remaining 22 oilimporting economies (rich, middle-income, and poor) are available but not reported due to space constraints. For US (Figure 8), an unanticipated oil supply disruption causes a small transitory decrease in crude import demand (negative and price inelastic) but petroleum import demand is not so responsive to this shock (literally zero price elasticity). A positive aggregate demand shock results in a strong and persistent increase in crude import than petroleum import with about a year delay (positive but price inelastic). Precautionary demand for oil causes a mild transitory increase in crude import demand but a nonnoticeable effect on petroleum import demand. Like the US, for Germany (Figure 9), an unanticipated oil supply shock causes a mild transitory decrease in crude import demand (negative and price inelastic) but petroleum import demand is not so responsive to this shock (literally zero price elasticity). A positive aggregate demand shock results in an immediate and persistent increase in crude import demand than petroleum import demand. However, compared to the US, Germany s petroleum import demand is more responsive to precautionary demand shocks. For the Netherlands (Figure 10), oil supply shocks do not have an immediate impact on raw crude and petroleum import demand as the price elasticity is zero in the first year and turns positive from about the second year. Crude and petroleum import demand elasticities in response to a positive aggregate demand shock is positive for most part of the five year horizon. Moreover, it appears that, the Netherland s immediate response to precautionary demand for crude and petroleum products is stronger relative to US and Germany. Regarding middle-income oil-importing economies, South Korea s and Thailand s raw crude import demand are driven primarily by aggregate demand shocks (Figure 10

11 11 and Figure 12), with Thailand s response being the strongest. However, Thailand appears to be less susceptible (fairly price inelastic) to adverse oil supply shocks both in its demand for raw crude and petroleum products, perhaps due to oil subsidies. Compared to Thailand, South Africa appears to be more susceptible to adverse oil supply shocks as evident in its combined crude and petroleum import demand (Figure 13). Additionally, the increase in South Africa s crude and petroleum import demand in response to a positive aggregate demand shock is not as strong as Thailand and South Korea. Precautionary demand for oil triggers an immediate increase in petroleum import demand in Thailand more than in South Korea and South Africa. Turning now to poor oil-importing economies, plots in Figure 14 (Bangladesh), Figure 15 (Kenya), and Figure 16 (El Salvador) indicate that Kenya is more susceptible to adverse oil supply shocks in their demand for raw crude import (negative and persistently price inelastic), while Bangladesh and El Salvador are less susceptible to adverse oil supply shocks in their demand for raw crude import (literally zero price elasticity). A common thread between El Salvador and Bangladesh is the existence of oil consumption subsidies which, like Thailand, may explain their relatively lower responsiveness to adverse oil supply shocks compared to Kenya. 2 A positive aggregate demand shock results in an immediate and strong increase in crude import in Kenya and El Salvador than in Bangladesh. Among the three poor oil-importing economies, import demand for petroleum products is more responsive to precautionary demand shocks in El Salvador than in Kenya and Bangladesh. 5 Conclusion The effects of higher oil prices on oil import demand remain a question of considerable interest in the theoretical and empirical literature on energy demand. The central message of the paper has been that the standard approach to modeling oil import demand as a function of oil price and GDP is not well-defined because it implicitly assumes that one can vary the price of oil while holding all other variables fixed. This ceteris paribus assumption is not appropriate for two reasons. First, the standard approach implies that oil import demand is endogenous while the price of oil and GDP are exogenous. But this is not the case since the price of oil and GDP are potentially endogenous and reverse causality may run from one to the other. Second, shocks to the global demand for all industrial commodities may have direct effects on oil import demand as well as an indirect effect working through the price of oil. Thus, even if one controls for the reverse causality, these direct and indirect effects of global demand shocks on oil import demand, invalidate the ceteris paribus assumption. The paper jointly addressed these issues using a structural VAR model. In particular, it identified the underlying demand and supply shocks in the global crude oil market and decomposed the real price of oil into three distinct components: oil supply shocks, aggregate demand shocks, and precautionary demand shocks. Consistent with the oil shocks literature, oil supply shocks are found to have historically made relatively small contributions to fluctuations in the real price of oil while the largest 11

12 contributions have come from aggregate demand shocks and precautionary demand shocks. The identification of these shocks was crucial not just for explaining fluctuations in the real price of oil but also for understanding the degree of responsiveness elasticities of oil import demand associated with oil price fluctuations. Relying on the panel ARDL approach to cointegration, the study estimated the price elasticities of import demand for crude and petroleum products to each of the decomposed oil price shock. The panel short run price elasticity estimates reveal that an adverse oil supply shock causes a mild transitory decrease in import demand for crude and petroleum products while an aggregate demand expansion causes a delayed but strong persistent increase in import demand for crude and petroleum products. A precautionary demand shock triggers an immediate increase in petroleum products import more than raw crude import. The country level elasticity estimates fairly reflect the panel estimates although there are subtle variations within and across rich and poor oil-importing countries. At a more general level, the analysis presented in this paper shows that, contrary to the standard modeling approach in existing oil import demand literature, crude and petroleum import demand do not depend on a single oil price shock but depend on multiple oil price shocks resulting from oil supply shocks, aggregate demand shocks, and precautionary demand shocks; and that the short run price elasticities to these three distinct oil price shocks vary both within and across oil-importing countries. Notes 1 The country classification is based on GNI per capita from Alternative classifications based on PPP GNI per capita and PPP GDP per capita (data on which is available from ) yielded similar country categorization. 2 According to IEA estimates for 2012, oil subsidies amounted to $100 million in El Salvador, $900 million in Bangladesh, and $2.9 billion in Thailand. Kenya has subsidies on electricity consumption but not on oil consumption (IMF, 2012). References Alquist, R., & Kilian, L What do we learn from the price of crude oil futures? Journal of Applied Econometrics 25 (4): Alquist R, Kilian L, Vigfusson RJ Forecasting the price of oil. In Handbook of Economic Forecasting, Vol. 2, ed. G Elliott, A Timmermann, Amsterdam: North-Holland Altinay, G Short-run and long-run elasticities of import demand for crude oil in Turkey. Energy Policy 35 (11): Asali, M Income and price elasticities and oil-saving technological changes in ARDL models of demand for oil in G7 and BRIC. OPEC Energy Review 35 (3): Barsky, R., & Kilian, L Oil and the Macroeconomy Since the 1970s (Working Paper No ). National Bureau of Economic Research. 12

13 Chang, S.C Effects of financial developments and income on energy consumption. International Review of Economics & Finance (35): Chen, Y., Yu, J. & Kelly, P Does the China factor matter: What drives the surge of world crude oil prices, The Social Science Journal 53 (1): Cooper, JCB Price elasticity of demand for crude oil: estimates for 23 countries. OPEC Review (27): 1-8. Dargay, J.M. & Gately, D World oil demands shift toward faster growing and less price-responsive products and regions. Energy Policy (38): Fattouh, B., Kilian, L., & Mahadeva, L The Role of Speculation in Oil Markets: What Have We Learned So Far? (CEPR Discussion Paper No. 8916). C.E.P.R. Discussion Papers. Atkins, F.J. & Jazayeri, S.M.T A literature review of demand studies in world oil markets Department of Economics, Discussion Paper University of Calgary, Calgary, AB Ghosh, S Import demand of crude oil and economic growth: Evidence from India. Energy Policy 37 (2): IEA s World Energy Outlook 2015 IMF World Economic Outlook. Tensions from the two-speed recovery: unemployment, commodities, and capital flows. Washington, DC, International Monetary Fund. Javan, A. & N. Zahran Dynamic Panel Data Approaches for Estimating Oil Demand Elasticity. OPEC Energy Review 39(1): Kao, C., Chiang, M.H., & Chen, B International R&D Spillovers: An Application of Estimation and Inference in Panel Cointegration. Oxford Bulletin of Economics and Statistics 61 (S1): Kilian, L. 2009a. Not All Oil Price Shocks Are Alike: Disentangling Demand and Supply Shocks in the Crude Oil Market. American Economic Review, 99 (3): Kilian, L Oil Price Shocks: Causes and Consequences. Annual Review of Resource Economics 6(1): Kilian L. & Murphy, D.P Why agnostic sign restrictions are not enough: understanding the dynamics of oil market VAR models. J. Eur. Econ. Assoc 10: Kilian, L. & Murphy, D. P The Role of Inventories and Speculative Trading in the Global Market for Crude Oil. Journal of Applied Econometrics 29 (3): Kilian, L., Rebucci, A. & Spatafora, N. 2009b. Oil Shocks and External Balances. Journal of International Economics 77 (3): Kim, H. S., & Baek, J Assessing dynamics of crude oil import demand in Korea. Economic Modelling 35:

14 Knittel, C. R., & Pindyck, R. S The Simple Economics of Commodity Price Speculation (Working Paper No ). National Bureau of Economic Research. Levin, A., Lin, C.-F., & James Chu, C.S Unit root tests in panel data: asymptotic and finite-sample properties. Journal of Econometrics 108 (1): Pedroni, P Critical Values for Cointegration Tests in Heterogeneous Panels with Multiple Regressors. Oxford Bulletin of Economics and Statistics 61 (S1): Pesaran, M. H General Diagnostic Tests for Cross Section Dependence in Panels (Cambridge Working Papers in Economics No. 0435). Faculty of Economics, University of Cambridge. Pesaran, M. H A simple panel unit root test in the presence of cross-section dependence. Journal of Applied Econometrics 22 (2): Pesaran, Shin, & Smith Pooled mean group estimation of dynamic heterogeneous panels. Journal of the American Statistical Association (94): Phillips, P. C. B., & Perron, P Testing for a unit root in time series regression. Biometrika 75 (2): Sun, W., Qi, Z., & Jia, N Import Demand of Crude Oil and Economic Growth in China: Evidence from the ARDL Model. In 2010 Third International Conference on Business Intelligence and Financial Engineering (BIFE): Alleyne, T Energy Subsidy Reform in Sub-Saharan Africa Experiences and Lessons. IMF, Washington D.C. von Arnim, R. & Prabheesh, KP Rebalancing through expenditure and price changes. International Review of Applied Economics 27 (4): Ziramba, E Price and income elasticities of crude oil import demand in South Africa: A cointegration analysis. Energy Policy 38(12): Tables and Figures 14

15 Table 1: Summary of the literature on elasticities of oil import demand Author Country/Period Methodology Oil type Short run price elasticity Short run income elasticity Cooper (2003) 23 Countries/ Nerlove s partial Crude oil From 9 to 1 - adjustment model Altinay (2007) Turkey/ ARDL cointegration Crude oil Dargay & Gately (2010) OECD, Income Growers, China, Oil Exporters, Former Soviet Union, Other Countries/ Pooled cross section Transport oil, Fuel oil, Other oil, Total oil From 0.03 to 0.04 From 0.16 to 0.88 Ziramba (2010) South Africa/ JJ cointegration, ECM Crude oil Sun et al. (2010) China/ ARDL cointegration Crude oil Asali (2011) G7 & BRIC/ ARDL cointegration Crude oil IMF (2011) 45 OECD & non- Panel data approach All oil From 0.01 to 0.03 From 0.67 to 0.71 OECD/ Kim & Baek (2013) South Korea/ ARDL cointegration Crude oil Javan & Zahran (2015) 25 Countries/ 1993 to 2012 Panel and pooled data regressions based on Nerlove s partial adjustment model Crude oil From to 0.07 From 0.01 to 1.09 Notes: ARDL: Autoregressive distributed lag, JJ: Johansen Juselius; Error correction model. Oil type definitions. Crude oil: raw crude oil. Transport Oil: Gasoline, Jet Fuel, Diesel (Light Fuel Oil used in transport); Fuel Oil: Residual Oil, Heating Oil (Light Fuel Oil not used in transport), Kerosene (non-jet fuel); Other Oil: Feedstock (petrochemical inputs: Naphtha and Liquefied Petroleum Gases, LPG), non-feedstock LPG, and Miscellaneous; Total oil = Transport oil + Fuel oil + Other oil. All oil: crude and petroleum products.

16 Figure 1: Percent change in monthly global crude oil production 1980:1-2013: Figure 2: Monthly global real economic activity index deflated with US CPI and detrented 1980:1-2013: Figure 3: Log monthly real oil price 1980:1-2013:

17 Figure 4: Historical evolution of the annualized structural shocks Oil supply shock Aggregate demand shock Precautionary demand shock

18 Table 2: Data definitions and sources Symbol Variable Definition and source PROD Oil production Monthly global crude oil production (in million barrels per day) REA Real economic activity Monthly index of real economic activity ROP Real oil price Monthly real oil price CRUDE-IMP Raw crude import Raw crude import (in million barrels per year) PETRO-IMP Petroleum import Petroleum products import (in million barrels per year) OIL-IMP Crude & petroleum Crude plus petroleum products import (in million barrels per year) NOP Nominal oil price Annual nominal Brent crude oil price CRUDE Crude import bill Cost of crude oil import (in billion dollars) PETRO Petroleum import bill Cost of petroleum products import (in billion dollars) OIL Oil import bill Cost of crude plus petroleum products import (in billion dollars) GDP Gross domestic product GDP (in current US dollars - billion) CRUDE IMP DD Crude import demand Crude import bill as a share of GDP PETRO IMP DD Petroleum import bill as a share of GDP OIL IMP DD Oil import bill as a share of GDP Notes: PROD is monthly global crude oil production from the US Energy Information Administration (EIA); REA denotes detrented monthly index of real economic activity a la Kilian (2009a) with most recent updates from Kilian's research webpage; and ROP denotes monthly real oil price derived based on refiner acquisition cost of imported crude oil (EIA) deflated by deflated by the US CPI. The sample period is 1980:1 2013:12. Percent changes in PROD and logs of REA and ROP were used to generate the structural shocks. Annualized shocks were then constructed as averages of the monthly structural shocks for each year. Annual crude and petroleum import data, in million tons of oil equivalent (mtoe), was collected from the International Energy Agency (IEA) for a sample of 31 oil importing countries ( ). A country is classified as oil importer if its average oil exports is less than 30% of its average oil imports during The import data is converted from mtoe to million barrels of oil per year (mbpy) to derive CRUDE-IMP, PETRO-IMP, and OIL-IMP. The conversion is based on Brent crude conversion factor of 1 ton of oil equivalent = 7.57 barrels of oil equivalent provided by Reuters. Nominal oil price (NOP) data was derived from BP Statistical Review of World Energy. Data on GDP (in current US dollars) was obtained from the IMF World Economic Outlook (WEO) database. CRUDE-IMP, PETRO-IMP, OIL-IMP, and NOP were used to generate the respective fuel import bills: CRUDE, PETRO, and OIL. Each fuel import bill was then expressed as a share of GDP to generate the response variables: CRUDE IMP DD, PETRO IMP DD, and OIL IMP DD.

19 Table 3: Summary Statistics Variables Mean Standard Dev. Min Max Obs. Monthly series PROD REA ROP Transformed monthly series used to generate structural innovations (shocks) ΔPROD LG-REA LG-ROP Data used to generate response variables CRUDE-IMP PETRO-IMP OIL-IMP NOP CRUDE PETRO OIL GDP Response variables CRUDE IMP DD PETRO IMP DD OIL IMP DD Notes: ΔPROD = percent change in monthly global crude oil production. LG-REA = log of monthly index of real economic activity. LG-ROP = log of monthly real oil price.

20 Table 4: Time series unit root test Augmented Dickey-Fuller Test Null: series not stationary Variable Test statistic Critical Value p-value ΔPROD LG-REA LG-ROP *** * Notes: ΔPROD is the percent change in monthly global crude oil production, LG-REA denotes the log monthly index of real economic activity and LG-ROP denotes the log of monthly real oil price. The sample period is 1980:1 2013:12. p-values: *** denotes 1% level of significance, ** 5%, and * 10%. Table 5: Cross-section dependence test Pesaran (2004) CD Test Null: cross-section independence Variable Test statistic p-value CRUDE IMP DD 88.33*** 0 PETRO IMP DD 66.69*** 0 OIL IMP DD 96.42*** 0 Notes: CRUDE IMP DD = Crude import bill as share of GDP. PETRO IMP DD = Petroleum products import bill as share of GDP. OIL IMP DD = Oil (crude plus petroleum products) import bill as share of GDP. p-values: *** denotes 1% level of significance, ** 5%, and * 10%. Table 6: Panel unit root test with cross-sectional dependence Pesaran (2007) CIPS Test Null: series not stationary Variable Test statistic p-value CRUDE IMP DD *** 0 PETRO IMP DD *** 0 OIL IMP DD *** 0 Notes: CRUDE IMP DD = Crude import bill as share of GDP. PETRO IMP DD = Petroleum products import bill as share of GDP. OIL IMP DD = Oil (crude plus petroleum products) import bill as share of GDP. p-values: *** denotes 1% level of significance, ** 5%, and * 10%.

21 Real price of oil Real price of oil Real price of oil Figure 5: Response of real price of oil to one-standard deviation structural shocks 15 Oil supply shock Aggregate demand shock Precautionary demand shock Months (Point estimates with 95% confidence interval)

22 Figure 6: Cumulative effect of oil demand and oil supply shocks on the real price of oil 1980:1-2013:12 Cumulative effect of oil supply shocks on real price of crude oil Cumulative effect of aggregate demand shocks on real price of crude oil Cumulative effect of precautionary demand shocks on real price of crude oil