Changes in Energy Intensity in Canada

Canada is one of the top energy users and CO2 emitters among the OECD countries. However, energy intensity has been declining, on average, by about 1.4 percent since 1980. In this paper, we use the Fisher Ideal Index to determine the contribution of changes in the composition of economic activities and efficiency to a decline in energy intensity in Canada at national, provincial, and industry levels. We also apply panel data estimation methods to further investigate the factors driving energy intensity, efficiency and activity indexes for the period 1981-2008. We test for endogeneity as well as cross-section dependency in the provincial data and control for factors such as climate, policy, and energy endowment. The national and provincial decomposition results suggest that most of the reduction in energy intensity has occurred mainly due to improvements in energy efficiency rather than shifts in economic activities. Within the industry, while energy intensity has declined significantly in manufacturing, it has remained stable in transportation, utilities, and construction, and increased significantly in oil extraction and mining industries. The provincial panel regression results indicate that energy intensity is higher in provinces with higher average incomes, faster population growth, colder climate, and a higher capital-labour ratio, and lower in provinces with higher energy prices and higher investment. The industry panel regression results show that investment has contributed to energy efficiency in utilities and mining and to a shift away from energy-intensive activities in manufacturing and transportation industries. Technological advances have been most effective in increasing energy efficiency in construction and utilities and in decreasing energy-intensive activities in manufacturing industries. The results indicate that although efficiency contributes to a reduction in energy intensity in Canada, increasing activity in energy-intensive industries, such as oil and mining, partially offsets the efficiency gains in other industries.


INTRODUCTION
Canada is one of the chief energy users in the world with its total energy use growing steadily since 1980.Although energy intensity, i.e., energy consumed per unit of output and measured by the ratio of energy consumption to GDP, has been declining in Canada recently, it is still 1.3 and 2.4 times greater than that in United States and Germany, respectively (Figures 1 and 2).It is important to note that a fall in energy intensity does not necessarily mean total energy consumption is falling.The ratio of energy per GDP can still fall even if total energy use rises because  particularly in the 1980s (Gardner, 1993; Howarth et al., 1993l, Boyd and Roop, 2004; Wing, 2007;  Metcalf, 2008, Natural Resources Canada, 2013).In the more recent studies, Huntington (2010)  and Mulder (2015) also highlight the importance of structural changes in the U.S. and some OECD countries due to international trade.Regression analysis is also used to explain the changes in energy intensity as a function of various socioeconomic variables.These studies report economic growth, FDI, and income as major determinants of changes in energy intensity (Gardner et al.,  1998; Fisher-Vanden et al., 2004; Wing, 2007; Metcalf, 2008; Zheng et al., 2011; Song and Zheng,  2012).
We contribute to existing literature by employing the Fisher Ideal Index to decompose the Canadian energy intensity at the two-and the three-digit NAICS industry level data for each industry.The decomposition method is also applied to provincial data comprising seven sectors.We further employ a panel data analysis including endogeneity and cross-section dependence tests to examine socioeconomic and climate factors that drive changes in energy intensity in Canadian provinces and industries.The findings of this study will shed more light on the dynamics of the changes in energy intensity in Canada, one of the highest energy-intensive countries, which can be utilized in making energy and environmental policies.

DECOMPOSITION METHOD
Energy intensity can be written as the weighted average of sectoral energy intensity, where weights are the output share of the sectors.That is, where e is energy intensity, E it and Y it are the total energy consumption and GDP for sector i in time t, respectively.Equation (1) indicates that the aggregate energy intensity is equal to the sum of the products of energy intensity within a particular sector (e it ) and changes in the structure of economic activity (s it ) across sectors. 2 The energy intensity index (I t ) is then constructed by dividing the energy intensity in year t (e t ) by the energy intensity in a base year (e 0 ): The energy intensity index can be decomposed into two factors: the efficiency index and the activity index.The efficiency index attributes energy intensity to efficiency change holding the economic activity constant, and the activity index attributes energy intensity to change in the mixture of economic activity keeping efficiency within a sector constant.The decomposition can be carried out by either the Laspeyres index, which uses a base period fixed weight, or the Paasche index, which uses an end period fixed weight as follows: Copyright ᭧ 2016 by the IAEE.All rights reserved.
3. The idea is based on Fisher's (1921) decomposition of an expenditure index into a price and quantity indexes (Metcalf,  2008).
4. Focusing on supply side, our analysis does not include the residential sector.These indexes produce different decompositions as they use different base years and the decomposed indices might not add up to the total energy intensity index.The Fisher Ideal Index is the weighted average of Laspeyres and Paasche Indexes, which perfectly decomposes energy intensity into two efficiency ( ) and activity ( ) elements with no residuals. 3That is, and the total energy intensity index can be written as a product of the two efficiency and activity indexes as follows: The energy savings can be allocated between efficiency and activity using the equation below: where E t is the actual energy consumption and is the actual energy that would have been consumed Ê had energy intensity remained at its base year level.

Data
The energy consumption and economic activities data are obtained from Canadian Socioeconomic Information Management System of Statistics Canada (CANSIM).The details of the data sources are presented in Table A1 in appendix.We first conduct the national level analysis using the two-digit NAICS level industry data for the period 1981-2008.Due to the inconsistency in the datasets, we regroup the data into 17 industries, a list of which is presented in Table A2 in Appendix. 4he industry classifications for economic activities have to match those for energy use in order to conduct the decomposition analysis.Thus, we use the real gross domestic product at industry levels for which the energy consumption data is available.As Fisher-Vanden et al. (2004)  note, decomposition using the aggregate data may generate misleading results as the likely changes in economic activities within a sector are not accounted for and are, therefore, ascribed to efficiency.Although decomposition using disaggregate data is more desirable, the exercise runs into the data availability problem.We, however, further construct a data set for some selected industries at the 5.We also examined two alternative base years (1990 and 2000) for our decomposition analysis, but the overall results remained unchanged.The results are available from the authors.
three-digit NAICS level for the period 1981-2008.This data set allows us to decompose the energy intensity index at the national level as well as at the selected industry levels.The list of the industries is presented in Table A3 in appendix.
We construct the provincial data for seven sectors: agriculture, mining and oil and gas extraction, construction, manufacturing, transportation, public administration, and services.Due to the lack of consistent provincial real GDP by industry that dates back to 1984, the real GDP is obtained by dividing the nominal GDP by an appropriate price index.Specifically, the provincial farm product index is used to obtain the real GDP for the agriculture sector, national IPPI (total excluding petroleum and coal product) for manufacturing, IPPI (petroleum and coal product) for oil, mining, and transportation, and provincial CPIs for all the other service sectors.

National Analysis
We first use the Fisher Ideal Index presented by equation ( 2) to decompose the energy index into efficiency and activity indexes using the two-digit NAICS industry data for the period 1981-2008.The decomposition results for Canada, taking 1981 as the base year, are illustrated in Figure 3. 5 Total energy intensity has declined by 26 percent between 1981 and 2008.Moreover, the activity index and the efficiency index were 90 percent and 82 percent of their 1981 level, respectively.That is, if energy efficiency had remained unchanged at its 1981 level for all sectors, energy intensity would have declined by 10 percent.Likewise, if composition of the economic activity had remained constant between 1981 and 2008, energy intensity would have declined by 18 percent.Note: Energy intensity is measured in terajoules per 1,000,000 dollars.Rows 2 and 3 assume the US and the EU economic structures for Canada.Source: CANSIM, EU-KLEMS, OECD Database, and authors' calculation Using Equation (3), we further compute the energy saved, assuming that energy intensity remained the same at its 1981 level and report the results in Figure 4. From 1981 to 2008, a total of 27.8 x 10 6 tera joules of energy or 13 percent of total energy use has been saved due to the decline in energy intensity.Improvement in efficiency accounted for 83 percent of the energy saved while changes in economic activity accounted for 17 percent.
Our decomposition analysis suggests that the contribution of structural change to the energy intensity decrease in Canada is 12 percent, which is lower than the 16 to 36 percent reported for the U.S. by Huntington (2010) and Metcalfe (2008), and the 25 to 42 percent reported for OECD-19 by Mulder (2015).The low share of structural change in the decline of energy intensity in Canada partly reflects a recent boost in the energy-intensive sectors of mining, oil, and gas extraction and transportation industries, which will be discussed in more detail in the next section.To further explore the contribution of structural changes to the decline of energy intensity in an international context, we calculate the changes in energy intensity in Canada using the corresponding economic structures of the U.S. and the EU countries.That is, we ask what the energy intensity in Canada would have been if the country had had the same economic structure as the U.S. and the EU-10 (western European) countries throughout the sample period.The results in Table 2 show that energy intensity in Canada would have fallen further by a factor of about 5 and 2, if the Canadian economic structure had remained the same as the U.S. and the EU-10, respectively.In other words,  with an economic structure similar to the U.S. and the EU-10, Canada would have been able to close its energy intensity gap with the rest of developed world and might even have decreased its energy intensity to a level much lower than what those countries currently have.

Industry Analysis
The industry level decomposition is carried out for the selected industries at the three-digit NAICS level for the period 1981 to 2008.The industries include manufacturing, transportation, mining, utilities, construction, and services.To construct the energy intensity, we use the gross domestic product (2002 constant prices) appropriate for each energy use sector for different industries.Table 3 shows the energy intensity indices in different industries, and Figure 5 demonstrates the changes in industry energy intensity index for the period 1981-2008.The more detailed results including the trends are shown in Table A4 in appendix.
Energy intensity in mining, oil and gas extraction has been increasing constantly throughout the period with an average growth rate of 1.23 percent per year; in 2008, it was 44 percent higher than its 1981 level.Changes in economic activity in the mining industry have been moderately stable mainly because the industry includes homogenous energy-intensive activities.The upward trend in energy intensity within this industry was mainly driven by the decline in energy efficiency with an average rate of 1.3 percent per year.Energy intensity in the utility industry has 6.Other sectors include wholesale and retail trade, utilities, information and cultural industries, education services, health care and social assistance and any other services not listed.We excluded the utility industry from this group, but the results did not alter.
7. Newfoundland and Labrador can be considered as an outlier because of a very high decline rate of energy intensity and efficiency index and a greater than one activity index.The former may be explained by very high energy intensity in the initial year, and the latter may reflect the structural change from fishing to oil and gas industry in the mid-1990s.
been stable until the mid-1990s, after which it started to increase, reaching its peak in 2001, when intensity was 134 percent of its 1980 level.Energy intensity has been declining mostly because of efficiency improvement since 2000, reaching its 1981 level in 2008.Energy intensity in the construction industry declined in the early 1980s and remained rather stable throughout the period.It has been declining with an annual rate of 0.7 percent reaching 84 percent of its 1981 level in 2008.Efficiency improvement has been the main source of declining energy intensity in the industry.
Energy intensity in the manufacturing industries has been declining on average, at an annual rate of 2 percent for the period 1981-2008, reaching 63 percent of its 1981 level in 2008.Improvement in efficiency has played a dominant role in this downward trend.Specifically, had energy efficiency remained unchanged, changes in economic activity would have reduced energy intensity to just 99.8 percent of its 1981 level.The manufacturing energy intensity trend shows that most of the improvements in the sector's energy intensity have occurred in the 1990s, because of increasing efficiency and shifts to less energy-intensive sectors such as information and communication technology industries.However, the efficiency improvement has continued at a much slower rate of 0.5 percent per year and the structure of industries shifted slowly to more energyintensive industries since 2000.Energy intensity in transportation industry has declined by 11 percent due to increasing efficiency mostly in the 1990s.The activity index indicates that the industry shifted steadily to more energy-intensive activities in the 1980s and the 1990s.Overall, if the efficiency had remained the same as its level of 1981, the energy intensity in transportation industry would have risen by 20 percent by 2008.Finally, energy intensity in services has declined by 44 percent in 2008, mostly due to improved efficiency.The changes in energy intensity in this industry have been smooth, with most of efficiency improvements taking place in the 2000s.

Provincial Analysis
Due to data limitation, we divide total energy use into seven sectors: agriculture, mining and oil and gas extraction, construction, manufacturing, transportation, public administration, and other sectors. 6To construct the energy intensity, we use the real gross domestic product corresponding to each energy use sector for the period 1984-2008.Since there is no consistent provincial real GDP by industry that dates back to 1984, we obtain real GDP by dividing the nominal GDP by appropriate price indexes.
Table 4 shows the decomposition results from Canadian provinces in 2008.Overall, all ten provinces have a downward trend in energy intensity, with Ontario, the main manufacturingbased province, having the lowest Intensity Index and Saskatchewan and Alberta, the two oil producing provinces, the highest.The variations in energy intensity across provinces are rather high and increasing over time, as the coefficients of variation have almost quadrupled between 1984 and 2008.As Table A5 in appendix shows, most of the decrease in provincial energy intensity has occurred since 1990.Specifically, Ontario experienced the greatest improvement in energy intensity with an average annual decline rate of 1.8 percent, and Saskatchewan and Alberta had the lowest average annual decline rates of about 0.5 percent. 7The gap between energy intensity in Saskatch-

Table 4: Provincial Energy Intensity and Decomposition Results (1984-2008)
Note: Energy intensity is measured in terajoules per 1,000,000 dollars.Source: CANSIM and authors' calculation ewan and Ontario has been widening since 1984, as the energy intensity in Saskatchewan was 27 percent higher than Ontario's in 1984, but increased to 81 percent in 2008.The last two columns of Table 4 show that both changes in the economic activity and energy efficiency improvement have played a role in reducing energy intensity in the provinces, but the impact of the latter has been much stronger than the former.Overall, Saskatchewan and Alberta have the highest activity index, which reflects the development of oil and mining industries in those provinces, particularly in the 1990s and 2000s, and Ontario and Quebec have the lowest efficiency index, which reflects the efficiency improvement in the manufacturing industries in those provinces.
The energy intensity indexes in Canada obtained from the provincial decomposition results are reported in the last row of Table 4. Comparing these figures with those from the industry analysis shown in Figure 3 indicates that the overall energy intensity indexes are similar.However, the contribution of efficiency is greater and of economic activity smaller in the provincial averages compared to the industry averages.This may be explained by the fact that the decomposition method using aggregate data overestimates the efficiency index because it does not allow for within-sector changes from high energy-intensive economic activities to low energy-intensive activities.In this case, any within-sector activity changes in the provincial sectors, such as manufacturing, would be attributed to efficiency rather than structural changes.
Table A6 in the appendix shows the total amount of energy saved throughout the 1984-2008 period.Because of a decline in energy intensity, all provinces, with the exception of Alberta, have experienced a reduction in energy consumption (saved energy).Efficiency improvement has led to energy savings in all provinces, except Alberta, and changes in economic structure has contributed to modest energy savings in four provinces, but higher energy consumption (negative saving) in six provinces.Ontario and British Columbia have significant energy savings due to efficiency improvement and a shift to less energy-intensive activities, but Alberta is the only province in which both changes in economic activity to more energy-intensive industries and a decline in energy efficiency increased energy consumption.This is mainly because of huge investments in the Alberta's oil sands, which is a high energy and capital-intensive industry.

SOCIOECONOMIC DRIVERS OF THE CHANGES IN THE ENERGY INTENSITY INDEXES
Although the decomposition method is an established tool to isolate the sources of energy intensity changes, it cannot explain the socioeconomic factors driving the energy intensity.In this 8.The partial adjustment model assumes that changes in y in each period is partially due to its deviation from its equilibrium level y*, so where represents the adjustment factor.Equation ( 5) is derived by substituting * Dy = γ(yy ), γ t t -1 equation ( 4) for y* and rearranging the terms.
section, we employ the regression analysis to analyze the underlying forces driving changes in the energy intensity.Our analysis includes estimation of the three energy intensity indexes using a panel data model with the provincial and industry data as follows: where y it is the intensity index for province i at time t, x it consists of socioeconomic variables, and is the idiosyncratic error term, u i is the province fixed effect controlling for the non-observable heterogeneities in provinces and industries, and v t is the time fixed effect controlling for macroeconomic and business cycles effects that affect all provinces or industries to the same degree.
The regression equation ( 4) assumes energy intensity responds contemporaneously to changes in economic variables.However, economic variables are likely to affect energy intensity with some lag because of timely capital and structural adjustments.A dynamic panel model can also be specified to control for the time adjustment required for the changes in energy intensity by using a partial adjustment model as follows: where is the short run and the long-run impacts of changes in x on y and 8 The h = γβ h/γ x = γ⑀ .
it it dynamic model allows us to estimate the short-run and long-run coefficients and elasticities.

Provincial Analysis
The specification of the energy intensity model is similar to that of the energy demand model; however, our dependent variable is energy intensity rather than energy consumption, which makes the interpretations of the coefficients different than the demand model.The socioeconomic variables included in our model are energy prices, income, capital-labour ratio, investment ratio, population growth, weather, and control variables for policy.By theory, higher energy prices should reduce energy intensity through the efficient use of energy and the shift from energy-intensive sectors.Under the assumption of a positively sloped aggregate supply due to sluggish wages in the short-run, higher prices may also reinforce the decline in energy intensity by increasing output.To capture the degree to which energy intensity responds to price changes in Canadian provinces, we use the provincial energy price index (2002 = 100), which includes electricity, natural gas, fuel oil , other fuels, gasoline, parts and supplies for recreational vehicles.
Income can have contrary effects on energy intensity.An increase in income may stimulate spending and a more energy-consuming lifestyle, which will lead to an increase in energy intensity.On the other hand, it can also increase environmental consciousness, which will lead to the adoption of energy-saving technology.It can be postulated that at lower levels of income, the former effect will dominate but as income rises, the latter effect will take over.Therefore, we use per capita income and its squared term to take into account the non-linear response of energy use to income.To control for the effect of weather on energy intensity, we use the heating degree days (HDD) and cooling degree days (CDD).Degree-days for a given day denote the number of Celsius degrees that the mean temperature is above or below a given base.For example, heating degree-days are the number of degrees below 18Њ C while cooling degree days are the number of degrees above 18Њ C. If the temperature is equal to or greater than 18, then the number of heating degrees will be zero.Values above or below the base of 18Њ C are used primarily to reflect the demand of energy required to cool or heat buildings.There are several weather stations in each province due to Canada's vast geography and varying weather.Our annual HDD and CDD data is constructed using the monthly data reported by weather stations within a province obtained from the Environment Canada database.
Increasing population can have a positive or negative effect on energy intensity.Fast growing provinces may add more energy efficient infrastructure than slow growing provinces.In contrast, if infrastructure does not keep up with growth, provinces growing quickly may be less energy efficient because of greater utilization of old and inefficient infrastructure and traffic congestion.We also use investment ratio as a proxy for the turn-over cycle of capital stock.Increasing investment may indicate the usage of improved and energy efficient capital which will lead to lower energy intensity.Thus provinces with higher investment are likely to have lower energy intensity than provinces with lower investment (Metcalf, 2008).The capital-labour ratio variable allows us to capture the technology effect on energy intensity.The capital-intensive technology may increase or decrease energy intensity depending on the relationship between capital and energy.Some studies suggest the relationship to be substitutionary while other studies postulate the relationship to be complimentary.Apostalakis (1990) notes that time series data tends to categorize capital and energy as complements because it reflects the short-term relationship.On the other hand, pooled crosssection studies capture the relationship as substitutes because it reflects the long-term relationship.Government energy policies and regulations can influence the energy consumption of its populace and consequently affect energy intensity.We use the reign of three distinct political parties (Conservatives, Liberal, and the New Democratic Party (NDP)) in Canadian provinces as a proxy for energy policy.Our model also includes a time fixed effect to capture the effects of technological progress and business cycles on energy intensity index over time.
The data sources are provided in Table A1 in the appendix and the provincial summary statistics in Table 5.The standard deviations show that there are variations both between and within provinces, but to different extents.For instance, energy price variations are greater within provinces over time than between provinces, but the variations in other variables are greater between provinces than within provinces over time.The average HDDs are twice as much as the CDDs with fewer variations over time and provinces, reflecting the cold climate of Canada.
We estimate a panel data model to explain the changes in energy intensity indexes across Canadian provinces for the period 1984-2008.The panel model specified in equation ( 5) is prone to some econometric issues including unobserved heterogeneity, endogeneity, and cross-section dependence.The unobserved heterogeneity among provinces and industries can be dealt with by the fixed effect estimation method. 9The endogeneity problem may arise with regard to the energy price and income variables.Energy prices, particularly oil prices, are determined in the international market and are, therefore, exogenous to the provinces and industries.However, electricity prices are likely determined by market forces locally, making the energy price index endogenous.Income and energy use may also be determined simultaneously as changes in energy use can be considered as a factor explaining changes in income.We apply the Durbin-Wu-Hausman method to test for endogeneity of the energy price and income variables, using the average energy prices of adjacent provinces and lag of income as instruments. 10We fail to reject null hypothesis of price and income exogeneity with the F test equal to 0.387 (p-value = 0.53) and 2.077 (p-value = 0.21), respectively, and the overall estimation results using the IVs remain unchanged.Finally, the fixed effect results above may be biased in the presence of cross-sectional correlation (Pesaran, 2004).Although Canada has a federal system with its provinces having a great deal of autonomy, it is still likely that the provinces respond similarly to common shocks implying that their economic performance and energy consumption are correlated.Thus, we conduct a test for cross-sectional correlation using the Frees and Friedman tests.The Frees' test value is 0.65 (critical value of alpha at 0.1 = 0.11), and the Friedman's test is 24.39 (Prob.= 0.004), indicating that the null of cross-sectional independency is strongly rejected by both tests.We, therefore, use the Driscoll-Kraay estimator (1998) which is robust to very general forms of cross-sectional and temporal dependence (Hoechle, 2007).
We run separate regressions for each energy intensity index obtained from the decomposition analysis.The energy intensity regression results show the effects of the covariates on the overall changes in energy intensity, whereas the efficiency (activity) index results show the effects of the covariates on energy intensity index should the economic structure (efficiency) remain unchanged.Table 6 reports the results of the fixed effects regression of the three energy intensity indexes in columns 1, 3, and 5 and the dynamic panel regression estimated by the Arellano and  Bond (1991) method in columns 2, 4, and 6.The static regression results can be interpreted as the long-run effect and the dynamic regression coefficients as the short-run effects. 11In general, the long-run effects are expected to be higher than the short-run effects because energy users will have enough time to adjust to the new conditions caused by changes in the socio-economic factors.Overall, the socioeconomic variables included in our model explain 80 percent of the variations of the intensity index, 74 percent of efficiency index, and 44 percent of activity index.Column (1) and ( 2) show that the coefficient of price is negative, but significant only in the long-run, and the income effect is positive and significant but slightly declining as income rises.The coefficient of capital-labour ratio is positive and significant, indicating that energy and capital are complementary in Canada.The investment ratio is negative but only significant in the short-run.The population growth coefficient shows that faster growing provinces have higher energy intensity.This could be because faster growing provinces suffer from congestion or attract energy-intensive infrastructure.The heating degree days (HDD) show a positive effect on energy intensity, but the cooling degree days (CDD) effect is positive only in the short-run.The reigns of the NDP and Liberal parties are associated with a small decline in energy intensity as compared to the Conservative party in the short-run.
Columns 3-6 report the estimation results for efficiency and activity indices.The results show that higher income, population growth, and HDD contribute to higher efficiency index, i.e., higher energy intensity, given that the initial economic structure remains the same.The reign of political parties, however, contributes to a slight improvement in efficiency index in the short-run.The activity index regression results also show that capital-labour ratio, HDD, and policies cause a shift to a higher energy-intensive activities in both the short-run and the long-run, but higher energy prices and investment ratio lead to less energy-intensive activities in the long-run.The positive coefficients for the political parties in the activity index might be surprising, particularly for NDP, which is known as a pro-regulation and pro-environment party and their reign is expected to be associated with lower energy-intensive activity.However, NDP is also a big supporter of unions, which have a strong presence in the energy-intensive industries such as manufacturing and mining.
Akin to the decomposition analysis, we can use the estimated coefficients to examine the effects of the individual factors driving the energy intensity using their initial levels.For instance, the energy intensity in 2008 would have been ten percent higher, if the energy prices had remained the same as their 1981 level.Likewise, if the income and other variables (investment ratio, popu- 12.Although income has a strong positive effect on energy intensity, its total effect including the squared income term is smaller in 2008 than 1981, making the 2008 energy intensity using the 1981 income levels larger.lation growth, HDD, and CDD) had assumed their 1981 levels, the 2008 energy intensity would have been higher by 29 percent and 2 percent, respectively. 12However, energy intensity would have been lower by 2 percent in 2008, if the capital-labour ratio had remained at its 1981 level, implying that employing more capital and energy-intensive technologies, particularly in the oil and mining industries, contributed to higher energy intensity.

Energy Endowment Effect
The regression in Table 5 assumes the homogeneous effects of the variables on energy intensity indexes in all provinces.Since the energy intensity in general is higher in oil-abundant countries, one might suspect that the energy intensity and its determinants in the oil-abundant Canadian provinces might also be different from those in the rest of country.We, therefore, run two separate regressions for the energy intensity in the oil-abundant provinces (Alberta, Saskatchewan, and Newfoundland and Labrador), and the rest of provinces.The results reported in Table 7 show that 89 percent of the variations of the energy intensity is explained by changes in socioeconomic variables.Specifically, energy prices are not significant in energy-endowed provinces, but significant in less energy-endowed provinces.Income is significant in both groups, but with a much stronger effect in less energy-endowed provinces.The coefficient of capital-labour ratio indicates that energy and capital are complementary in energy-endowed provinces, but are substitutes in the less energy-endowed provinces, which can be explained by the fact that energy-endowed provinces use more energy-intensive capital.Higher population growth increases energy intensity in both 13.During the 1984-2008 period and in 10 provinces, NDP has been in power 45 province-year (20%), Liberal party 80 province-year (34%), and Conservative party 108 province-year (46%).About 35% of the NDP province-year has been in Saskatchewan, but liberal and conservatives' province-years have been more evenly distributed across the provinces, so the reign of long serving Conservatives in Alberta does not have the same effect as that of NDP in Saskatchewan.
14. R&D investment may be a better proxy for technological advances, but the data is not long enough for having a meaningful estimation at the industry level.
groups but with a much stronger effect in energy-endowed provinces, implying investment in high energy-intensive infrastructures in those provinces.The coefficient of heating degree days is positive and significant in both groups, but it is higher in energy-endowed provinces, reflecting the fact that two main oil-rich provinces (Saskatchewan and Alberta) have the coldest temperatures in Canada.The policies of NDP governments have led to a significant increase in energy intensity in energyendowed provinces, but to a decline in less-energy-endowed provinces, which may be explained by the fact that despite the environmentally friendly agenda of the party, the long serving NDP government in Saskatchewan encouraged investment in its oil fields to boost economic growth of the province during the period 1990-2006.The reign of the Liberal party is associated with higher energy intensity in energy endowed provinces and lower energy intensity in less energy endowed provinces in the short-run. 13

Alternative Energy Prices
Our findings that energy prices do not significantly affect energy intensity may be driven by the aggregate energy price we use.As a robustness check, we consider two variations for the energy prices.First, we use relative energy prices measured by the ratio of energy prices to nonenergy prices, and second, we disaggregate the energy prices into electricity and natural gas prices.The overall regression results with the relative energy prices, which are not shown here, remain the same.Since our aggregate measure of energy prices might hide the true effects of different energy prices in provinces, we also re-estimate the model using the provincial electricity and natural gas prices.Although electricity price data are available for all provinces, the natural gas price data are not available for Maritime Provinces (NF, PEI, NB, and NS) leaving us a smaller sample size of 146.To avoid the reduction in sample size, we also run a regression assuming that the natural gas prices in Maritime Provinces are the same as those in the closest province, i.e., Quebec.The full sample results reported in Table 8 show that electricity prices are not significant, but natural gas prices are negative and significant, particularly in the long-run.The results with respect to the other coefficients remain almost the same.

Industry Analysis
We use the three-digit NAICS data to estimate the energy intensity indexes for Canadian industries.The industries included in the estimation are manufacturing, mining, construction, utility, transportation, and services.The estimation model includes energy prices, capital-labour ratio, investment ratio, and TFP growth as a proxy for technological changes. 14The results are reported in Table 9. Columns 1 and 2 show that the investment ratio and TFP growth have negative and significant impact on energy intensity, but a higher capital-labour ratio increases the energy intensity, implying that capital and energy are complementary.The energy prices have a negative, but not significant, effect in both the short-run and the long-run.The estimated coefficients in the efficiency index regression are almost the same as those in the energy intensity index, but none of the coef-  ficients in the activity index regression is significant, implying that changes in efficiency has been the main driver in changes in energy intensity, consistent with our findings in the decomposition analysis.
To shed more light on the individual industry effects, we re-estimate the industry models allowing for heterogeneity in the major variable effects such as energy prices, capital-labour ratio, investment ratio, and the TFP growth.The results presented in Table 10 show that a higher capitallabour ratio in mining and manufacturing industries has pushed the energy intensity upward, and TFP growth in utilities and transportation and higher investment ratio in utilities and mining have improved energy intensity.The efficiency index regression results also show that the significant improvement in efficiency index can be ascribed to higher energy prices in manufacturing in the long-run, higher investment ratio in mining and utilities, and higher TFP growth in utilities.However, employing more energy-intensive capital has deteriorated the efficiency index in mining and oil and gas extraction industries.The activity index regression results also reveal that changes to less energy-intensive industries are due to a higher investment ratio in manufacturing and transportation, higher TFP growth in manufacturing, and a higher capital-labour ratio in construction.

Table 11: Price and Income Elasticities
In the long-run, energy prices induce a shift to less energy intensive activities in manufacturing, utilities, mining and oil extraction, and services, and a higher capital-labour ratio in construction.

Elasticities
Using the estimated coefficients from the energy intensity index regression equations, we can obtain price and income elasticities for energy demand, which are and , h /I (h + h lny )/I + 1 respectively.Table 11 shows the short-run and long-run elasticities for energy demand using the estimated coefficients from the dynamic regression equations evaluated at the average values for energy intensity index and log of per capita income.The elasticities obtained from the energy intensity regressions are interpreted similar to those estimated from the energy demand equations.However, the estimated elasticities from the efficiency index and the activity index regressions should be viewed as responses to energy demand when economic structure and energy efficiency are kept constant, respectively.The price elasticities are negative for the intensity index and activity index and positive for the efficiency index, but none is significant.The income elasticities are all positive and significant (except for efficiency in the long-run).The greater-than-one income elasticities for the activity index indicate that higher energy demand is mostly driven by energy-intensive activities, such as mining and oil and gas extraction.We also estimate separate elasticities for two energy-endowed and less energy-endowed provinces.The price elasticities are negative and significant for intensity 15.Bruneau (2014) also reports no significant response of Canadian industries trade to electricity prices, but significant effect of natural gas prices.To save space, the regression results and elasticities for electricity and natural gas prices for energy-endowed and less energy-endowed provinces are not reported here, but they are available upon request.
16.The electricity price elasticities reported by Bernstein et al. (2005) are in the range of -0.01 and -0.28 for the short-run and -0.05 and -0.3 for the long-run.The energy price elasticities reported by Metcalf (2008) are in the range of -0.018 and -0.105 in the short-run and -0.056 and -0.299 in the long-run.David and Razek (2012) also report nonsignificant short-run price elasticities in Canada ( -0.01 to -0.02), but significant long-run elasticities ( -1.6 to -3.7).The income elasticities reported by Metcalf (2008) are in the range of -0.289 to 0.160 in the short-run and -0.820 to 0.524 in the long-run.and efficiency indexes in less energy-endowed provinces, but positive and significant for efficiency index in energy-endowed provinces.The income elasticities are positive and significant in both groups, and they are greater for the activity index, particularly in energy-endowed provinces.The results of the disaggregated energy prices show a positive electricity price elasticity for energy intensity and efficiency indexes and a negative elasticity for the activity index in both groups.However, the electricity price elasticities are much smaller in less energy-endowed provinces, most of which generate electricity using hydro or nuclear plants and have significant excess capacity. 15lthough our estimated price elasticities are small, they are within the ranges reported by Bernstein  et al. (2005) and Metcalf (2008) for the U.S. and David and Razek (2012) for Canada. 16Our income elasticities for energy intensity and efficiency indexes are also within the range, but for activity index are higher than those estimated by Metcalf (2008).

CONCLUSION
This paper provides a comprehensive analysis of the forces driving changes in energy intensity in Canada since 1980.We employ the Fisher Ideal Index to perfectly decompose energy intensity at the national, provincial, and industrial levels, and use econometric methods to identify underlying factors driving the changes in energy intensity in Canada.
The decomposition results, both at the national and provincial levels, suggest that most of the reduction in energy intensity has occurred in the 1990s and mainly due to improvements in energy efficiency, not to shifts from energy-intensive to less energy-intensive economic activities.Specifically, energy efficiency improvement accounted for more than 82 percent of the overall decline in energy intensity.Additionally, variation in energy intensity across provinces has been increasing over time.Within the industries, while energy intensity increased significantly in mining and oil extraction industries, it experienced a significant decline in manufacturing mostly due to an improvement in efficiency.The energy intensity has remained rather stable in other industries.
The panel data regression results also indicate that, on average, higher energy prices have led Canadian economic structure to move away from energy intensive activities, while rising income has been the most significant factor in increasing energy intensity.Even though population growth is relatively low in Canada, it has a positive and significant effect on energy intensity.Energy intensity increases with a decline in temperature, but the effect of a warmer climate on energy intensity is relatively limited.The provincial and industrial level study shows that increased use of energy-intensive capital has raised energy intensity, implying that capital and energy are complementary on average across provinces and industries.The investment ratio, which captures the turnover of capital stock, has also contributed to the decline in energy intensity in provinces.The industry regression results also confirm the investment effect and show that it has contributed to energy efficiency in utilities and mining and to changes to less energy-intensive activities in manu-facturing and transportation industries.Technological advances have been most effective in increasing energy efficiency in construction and utilities and in switching to less energy-intensive activities in manufacturing industries.The regression analysis for the two energy-endowed and the less energy-endowed provinces also reveals heterogeneous responses of energy intensity indexes to explanatory variables.Specifically, energy prices and income have stronger negative and positive effects, respectively, in less energy-endowed provinces.Also, policy effects are different in the two groups with Liberals having increased energy intensity in energy-endowed provinces and decreased it in less energy-endowed provinces.The energy demand elasticities results indicate that energy is price inelastic and changes in energy prices will reduce energy demand only in less energy-endowed provinces.However, breaking down the energy prices into electricity and natural gas prices in the regression reveals that while all provinces respond significantly to changes in natural gas prices, the electricity price elasticity is significant only in the less energy-endowed provinces.Furthermore, a rise in income will increase energy demand mostly due to a rise in high energy-intensive activities particularly in energy-endowed provinces.
Our study shows that Canada is slowly reducing its high energy intensity with a focus on increasing energy efficiency through economic forces such as investment and technological advances.However, increasing activity in energy-intensive sectors, such as oil and mining and transportation, partially offsets the efficiency gains in these and in other industries.This is particularly true as about 50 percent of the greenhouse gas produced in Canada is concentrated in oil and gas and transportation industries and in the two oil producing provinces: Saskatchewan and Alberta.Thus, the pace of energy intensity reduction will increase rapidly, should efficiency improve significantly in the energy-intensive industries, or provinces move to less energy-intensive activities.Since the latter is not a realistic option for Canada as a major oil-exporting country, the government policy to encourage R&D in those energy-intensive industries will help meet the CO 2 reduction targets in due course.Note: changes in sub-periods are average changes and in the full period (1981-2008) is the difference between the first and the last year.

Table A5: Energy Intensity Changes in Canadian Provinces (1984-2008)
Note: changes in sub-periods are average changes and in the full period (1981-2008) is the difference between the first and the last year.

Figure 3 :
Figure 3: Energy Intensity Indexes in Canada (Two-digit Industry Level)

Figure 4 :
Figure 4: Energy Savings Due to a Declining Energy Intensity in Canada (1981-2008)

Figure 5 :
Figure 5: Changes in the Energy Intensity Index in Canadian Industries (1981-2008)

Table 6 :
Provincial Regression Results (1984-2008) Notes: Dependent variables are energy intensity, efficiency, and activity indexes.All variables (except for energy index) are in log form.The fixed effect estimation results with the Driscoll-Kraay standard errors are in columns 1, 3, and 5, and the dynamic regressions estimated by the Arellano-Bond estimator with province and time fixed effects are in columns 2, 4, and 6. * pϽ0.10, ** pϽ0.05, and *** pϽ0.01.

Table 7 :
Energy Endowment Effect (1984-2008) Notes: Dependent variable is the energy intensity index.All variables (except for energy index) are in log form.The fixed effect estimation results with the Driscoll-Kraay standard errors are in columns 1 and 3, and the dynamic regressions estimated by the Arellano-Bond estimator with province and time fixed effects are in columns 2 and 4. * pϽ0.10, ** pϽ0.05, and *** pϽ0.01.

Table 8 :
Provincial Regression Results with Individual Energy Prices (1984-2008) Notes: Dependent variables are the energy intensity, efficiency, and activity indexes.All variables (except for energy index) are in log form.The fixed effect estimation results with the Driscoll-Kraay standard errors are in columns 1,3, and 5, and the dynamic regressions estimated by the Arellano-Bond estimator with province and time fixed effects are in columns 2,4, and 6. * pϽ0.10, ** pϽ0.05, and *** pϽ0.01.

Table 9 :
Energy Intensity Estimation Results for Canadian Industries (1981-2008) Notes: Notes: Dependent variables are energy intensity, efficiency, and activity indexes.All variables (except for energy index) are in log form.The fixed effect estimation results with the Driscoll-Kraay standard errors are in columns 1,3, and 5, and the dynamic regressions estimated by the Arellano-Bond estimator with province and time fixed effects are in columns 2,4, and 6. .* pϽ0.10, ** pϽ0.05, and *** pϽ0.01.

Table 10 :
Estimation Results for Industries with Industry Specific Coefficients (1981-2008) Notes: Notes: Dependent variable are energy intensity, efficiency, and activity indexes.All variables (except for energy index) are in log form.The fixed effect estimation results with the Driscoll-Kraay standard errors are in columns 1,3, and 5, and the dynamic regressions estimated by the Arellano-Bond estimator with province and time fixed effects are in columns 2,4, and 6. * pϽ0.10, ** pϽ0.05, and *** pϽ0.01.

Table 3 : The Industry Energy Intensity and Decomposition Results (1981-2008)
Note: Energy intensity is measured in terajoules per 1,000,000 dollars.Source: CANSIM and authors' calculation

Table A2 : The 2-digit NAICS Industries for the Decomposition Analysis
Source: CANASIM, Statistics Canada (see TableA1 in appendix)

Table A3 : The 3-digit NAICS Industries for the Decomposition Analysis
Source: CANSIN, Statistics Canada (see Table A1 in appendix)