Review on Energy Resilience

Economic Development and End-Use Energy Demand Author(s): Kenneth B. Medlock III and Ronald Soligo Source: The Energy Journal, Vol. 22, No. 2 (2001), pp. 77-105 Published by: International Association for Energy Economics Stable URL: https://www.jstor.org/stable/41322914 Accessed: 01-11-2019 18:11 UTC

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide

range of content in a trusted digital archive. We use information technology and tools to increase productivity and

facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at

https://about.jstor.org/terms

International Association for Energy Economics is collaborating with JSTOR to digitize, preserve and extend access to The Energy Journal

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

Economic Development and End-Use Energy Demand

Kenneth B. Medlock ///*, and Ronald Soligo**

We examine the relationship between economic development and energy demand. The paper identifies the development patterns that characterize particular economic sectors t and analyzes the effect of sector-specific energy demand growth rates on the composition of final energy demand. We also examine some of the associated policy implications. Industrial energy demand increases most rapidly at the initial stages of development, but growth slows steadily throughout the industrialization process. Energy demand for transportation rises steadily ' and takes the majority share of total energy use at the latter stages of development. Energy demand originating from the residential and commercial sector also increases to surpass industrial demand, but long term growth is not as pronounced as it is in the transport sector. These results have implications for the primary energy demand of an economy as it develops, and thus, for domestic energy security and global geopolitical relationships.

INTRODUCTION

The economic development of nations results in increased demand for energy resources. Shifts in the structure of consumption and production, however, alter the impact that changes in output have on changes in energy demand. This paper identifies general tendencies of end-use energy demand by sector (which, for our purposes, are defined as residential and commercial, industrial and other, and transportation) that are characteristic of the development process. Using panel data consisting of 28 countries from all levels

The Energy Journal, Vol. 22, No. 2. Copyright ° 2001 by the IAEE. All rights reserved.

Funding for this research was provided by the Center for International Political Economy. We thank participants of the 21st Annual Conference of the IAEE (Rome, Italy, June 1999) and three anonymous referees for helpful comments. As usual, any errors are our own.

* Baker Institute Scholar, James A Baker III Institute for Public Policy, Rice University, Dept. of Economics - MS 22, PO Box 1892, Houston, TX 77251-1892. E-mail: medlockOrice.edu

** Professor of Economics, Rice University, Department of Economics - MS 22, PO Box 1892, Houston, TX 77251-1892. E-mail: [email protected]

77

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

78 I The Energy Journal

development, we construct a 'map' of energy use by sector during the course of economic development.1 We use this development 'map' to project possible future growth of energy demand by end-use sector, and determine the composition of total final energy consumption as a function of the level of development. In addition, by analyzing energy use trends by sector, we are able to identify each sector's contribution to aggregate energy intensity at each level of per capita income. This, in turn, allows us to identify an aggregate energy intensity curve that has an upside-down "U-shape" and is asymmetric about its peak.

The formation of accurate expectations about the future growth and composition of energy demand in developing nations is important when considering the energy-related political and environmental pressures that may be faced in the future. Moreover, accurate expectations allow for the development of realistic and achievable energy security and environmental policies. For example, the OECD, which represents only about one-fifth of the world's population, accounts for well over half of global energy consumption. Industrialization in less developed countries (LDCs), however, will shift the share of energy consumption away from OECD countries. As the energy requirement of the LDCs expands, therefore, there will be increasing pressure to expand existing resource bases and increase the efficiency of end-uses in order to maintain relatively cheap fossil fuels. In 1998 the burning of fossil fuels accounted for 85% of global energy use, a fact that will most likely persist for the foreseeable future. The dependence on fossil fuels and the consequent release of carbon dioxide places an emphasis on the development of cleaner alternative energy resources (such as solar or fuel cell) and reducing carbon intensity. Energy policy, therefore, must give consideration to the future development of LDCs with regard to domestic energy security and the potential costs of environmental degradation.

II. STRUCTURAL DETERMINANTS OF ENERGY DEMAND

In this study we only examine commercial energy use. It is recognized that traditional energy resources, or combustible renewables and waste, are used to varying degrees in LDCs, but their use as a proportion of total energy declines significantly as commercial fuels become affordable. In addition, for the countries in our sample, combustible renewables and waste are consumed primarily in the residential sector. The percentage of combustible renewables and waste in total primary energy is given in Table 1 for a few of the LDCs in our sample. Combustible renewables and waste accounted for a smaller

1. Other panel data studies of note are Galli (1998), which estimates the effect of income and price on total energy use, and Judson et al. (1999), which estimates the effect of income (absent price) on energy use by sector. Our study can be viewed as a sort of hybrid of those studies.

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

Economie Development and End-Use Energy / 79

proportion of primary energy use in 1995 than in 1985 in every case. Moreover, the residential sector is where most of all traditional fuel use occurred. This

latter point is important when considering the empirical results below. Omitting traditional fuels in our analysis may present some bias,

particularly in the residential and commercial sector estimates. Total energy use will rise faster than indicated by our estimates in those sectors at the initial stages of economic development, but this growth will be due to the consumption of traditional fuels, not commercial energy. This bias will not affect an assessment of the impact of economic growth on the international energy market and commercially traded energy resources, however, because traditional fuels are generally not traded whereas commercial energy is.

Table 1. Traditional Fuel Use in Selected LDCs

Traditional Fuel in Primary Energy % of Traditional Fuel Consumed in the Residential Sector

Country 1985 1995 1995

India 55.5% 43.2% 88.6%

China 26.7% 19.5% 100.0% Indonesia 50.9% 35.2% 100.0% Pakistan 49.3% 40.6% 88.9%

Malaysia 10.7% 6.7% 95.4% Thailand 40.4% 28.6% 64.4%

Source: IEA, Energy Balances ofNon-OECD Countries

Changes in the structure of production and consumption that occur as economic development progresses are important factors in determining growth of energy demand. The relationship between economic development and the structure of production is one that is well-documented (see, for example, Kuznets, 1971; and Chenery and Syrquin, 1975). In the initial stages of economic growth, the share of industry in total output rises and the share of agriculture declines. As a result, the share of total energy use in the industrial sector is quite high in the initial phases of development. In the latter stages of development, however, the share of services increases to a point where it dominates total output. Rising incomes generate new demands from the consumer sectors, and, consequently, the shares of total energy use in the residential and commercial and transport sectors increase.

The industrialization process is typified by large increases in energy consumption. For example, the production of cement and steel for the development of infrastructure requires much more energy input per unit output than traditional methods of agriculture. Accordingly, the energy intensity of GDP initially increases (where energy intensity is defined as energy unit input per unit output). Two things happen, however, as consumer wealth rises. One, as incomes rise, an increasing share of the consumer budget is devoted to

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

80 / The Energy Journal

manufactured goods. In response, the industrial structure of a developing economy will begin to change, with a larger emphasis being placed on the production of consumer items (light industry).2 Two, there is an increase in the demand for service activities, which are typically less energy intensive than manufacturing. The combined effect of the transition within the manufacturing sector to light industry with the growth of the service sector is a reduction in the energy intensity of GDP.3 This phenomenon is often referred to as dematerialization, the process by which material inputs into production decrease per unit output (see Bernardini and Galli, 1993, for more on this issue).

The cross-section evidence in Figure 1 indicates the influence of changing economic structure on aggregate energy intensity. As countries move from low to lower-middle income status, there is an increase in the average energy intensity. However, the average energy intensity decreases when moving from upper-middle to high-income status.4 On average, the share of agriculture declines from 38% to 3% and the share of services rises from 40% to 67%. The

share of industry increases from 22% to 34% before falling to 30%. The pattern of increasing then decreasing energy intensity across stages

of development arises in large part because overall energy intensity is a weighted-average of the energy intensity within each sector. Aggregate energy intensity increases at the initial stages of development as the industrial sector grows relative to other sectors, and reverses itself as the less energy intensive service sector grows relative to agriculture and industry in the latter stages of development. It is important to note that declining energy intensity does not imply declining energy demand, only that energy demand grows more slowly than output.

As per capital income rises, consumer activity accounts for an increasing proportion of total energy demand. Consumer durables (such as air conditioners, furnaces, refrigerators, and automobiles) take up an increasing share of the consumer's budget (World Bank Development Indicators, 1997), and the utilization of these items increases energy requirements in the residential and commercial and transportation sectors. As households become saturated with energy consuming durable goods, however, the per capita growth rate of the energy demand in those sectors will begin to fall below the growth rate of per capita income. For example, thermostatic readings on air conditioners will invariably reach some comfort level for the individual and will not be changed. Thus, even if operated 24 hours a day, the energy consumed will reach some upper bound. In the case of transportation, automobile usage faces the same sort of restrictions, with time ultimately limiting the amount of driving one can do.

2. There is not perfect correlation to the degree that nations engage in international trade. 3. Innovations and improvements in energy efficient technologies will reinforce this trend. 4. Note also that these trends generally hold within countries over time as well as across

countries illustrated here, but for the purpose of exposition the cross-section will suffice.

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

Economie Development and End-Use Energy / 81

1-H

I

I

I I

I

11

о- р

§ о 00 I/)

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

82 I The Energy Journal

This "saturation effect," therefore, contributes to the trend of declining energy intensity with increasing income.

A factor that can potentially offset the saturation effect in the residential

and commercial sector is the expansion of floor space. While the consumption of durable services is constrained by both income and time, the expansion of floor space faces primarily an income constraint. Thus, as incomes rise, individuals can afford bigger homes. More square footage, ceteris paribus, requires more energy for heating and/or cooling. However, diminishing marginal utility to additional square footage implies that floor space expansion will eventually slow. In estimating energy consumption by sector below, we choose a functional form that will allow for saturation, should its effects be present, but does not arbitrarily impose it. This effectively enables us to test the 'saturation' hypothesis.

Energy intensity first increases then decreases through the development process. Growth in total energy demand will reflect the changing energy intensity in each end use, which is itself reflective of the changing structure of production and consumption in an economy. In particular, the income elasticity of energy demand should decline as countries move beyond the industrialization phase of development.

III. ANALYZING ENERGY DEMAND

The relationship between per capita energy demand and economic development has been explored extensively. Until recently, empirical studies have focused on global development patterns through the use of cross-section data (see, among others, Zilberfarb and Adams, 1981; or Ang, 1987), or individual country development patterns through the use of time-series data (see, for example, Pourgerami and Hirschhausen, 1991). In the absence of sufficient time-series data, the cross-section approach can be used to identify the long-run effect of income on energy demand, often called the energy coefficient. The cross-section approach, however, suffers from implicitly assuming that the same regularities apply to all nations. Countries can differ in their energy use because of factors such as different technologies, different energy prices (due to things such as different energy taxes), different factor endowments, different climates and so forth. Ignoring these factors may bias the estimated energy coefficient.

The time-series approach enables the researcher to specifically account for any factors that may be specific to an individual country, such as climate or demography, thereby avoiding potential problems of country specific heterogeneity. However, difficulties in obtaining reliable data sets of sufficient length, especially for LDCs has resulted in a lack of research into the energy demand patterns for developing economies. In some cases data constraints have prompted the omission of energy prices, which can be problematic because the long run effect of higher prices may be reductions in energy requirements per unit of output (either through substitution to alternative inputs or less energy

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

Economie Development and End-Use Energy / 83

intensive production methods, or through increased efficiency). The omission of prices, therefore, presents a potentially serious mispecification problem.

Many empirical studies, regardless of the approach taken, specify constant elasticity (log-linear) demand equations, which is not consistent with a large body of research. In particular, the estimated income elasticity for energy demand is generally lower for countries with levels of GDP per capita. The inability of log-linear specifications to account for this has been attributed to a problem of omitted variables, such as changes in capital stock efficiency (see, for example, Prosser, 1985), rather than as a fundamental characteristic of the structural relationship between energy demand and GDP.

Using cross-section data for 22 countries from 1950 to 1965, Brookes (1973) estimated a log-log relationship between per capita income and per capita energy use. He first estimated the entire set of nations year by year, then the divided sample, developed and less developed, also year by year. He found that the estimated income elasticity of energy demand fell over time for the entire sample. He also found that the income elasticity for the less developed countries was consistently higher than for the developed countries. Brookes attributed his findings to structural differences across the stages of development, citing that "post-industrial" development could lead to significant reductions in the income elasticity of energy demand.

Brookes' findings support the hypothesis that there is some non- monotonic relationship between per capita energy consumption and per capita GDP. Structural changes in production and consumption alter the impact of future economic growth on energy demand, but this effect is not evident in log- linear models. Hence, any such estimated relationship for a low-income nation is likely to change significantly as that nation grows economically, and, as a result, inferences based on log-linear analyses may be misleading.

Recent studies by Galli (1998) and Judson, Schmalensee and Stoker (1999) (henceforth JSS) are two examples of the more innovative recent work that supports the notion that the income elasticity of energy demand declines as income rises. Galli (1998) analyzes the energy-income relationship for a panel of ten Asian countries and finds that energy intensities tend to fall beyond some threshold level of income, which is estimated to be, on average, $3945.44. It is argued, therefore, that energy forecasting models must accommodate the existence of such a turning point less they suffer from significant mispecification bias. Galli 's study is limited to a sample of ten developing Asian countries, the most advanced of which is Korea, covering the period from 1973-1990. Our study includes a sample of 28 countries with greater representation of developed countries, and covers the period 1978-1995. We also expand our scope to include countries outside developing Asia. By including a larger sample of developed countries along with LDCs, our study gives a more complete picture of the energy-income relationship.

JSS disaggregate energy consumption by sector, and analyze energy demand patterns for a panel of 123 nations from all levels of development. Their

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

84 / The Energy Journal

findings indicate that the share of household energy consumption in total falls with increasing income, the share of energy consumed in transportation activities rises, and the share of industrial energy consumption follows an inverted U- shape pattern. JSS do not account for the effects of energy prices within their study, a factor attributed to a lack of data availability. The inclusion of prices would allow for a distinction between the impact of dematerialization (an income effect) and the impacts of changes in the relative price of energy. Given the rather volatile behavior of world energy markets over the past two decades, the

possible impact of energy prices on energy demand is something that should not be neglected. Doing so will inevitably disguise the true effect of income. For example, a decline in energy consumption following any major shock to energy prices might take a number of years to take full effect (assuming there is some lag in adjustment to equilibrium). Coinciding with increasing levels of development, the long-run effect of price increases could incorrectly amplify the downward effect that increasing incomes have on energy demand growth rates. Thus, the influence of dematerialization would be overstated. While there are advantages to estimating relationships with a large number of observations, such as the data set used in JSS, the estimated parameters will invariably suffer from some bias due to omitted variables. Weighing the suspected size of that bias against the advantage of expansive data sets, we sacrifice countries for price data, a procedure that considerably reduces the size of the sample.

IV. THE MODEL

The research by Galli and JSS is distinguished from that of others because they specify a relationship between energy and income that allows the income elasticity of demand for energy to change with income. Indeed, given the preceding arguments, there is no reason to believe that energy demand growth is a constant proportion of GDP growth. As a result, it is important to derive an estimable relationship that is flexible enough to permit this.

Since we are concerned with patterns of energy demand growth over the course of development, we seek a relationship that specifies those patterns. We assert that energy consumption, ect , is a function of per capita output, yn a vector of energy prices, pn and technology т„ or

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

Economie Development and End-Use Energy / 85

< =f(yrPrTt)

where the star denotes long-run optimality.5 Furthermore, ex post total final energy consumption is the sum of final consumption in each end-use

ectm = £ ectj j

where y denotes an end-use sector. For our purposes,./ represents residential and commercial, transportation, or industrial and other.

Assuming that technology is a function of energy prices and common across countries for a given level of economic development, we can define the

function / such that

ec,j =f{yt,Ptj^(yrPtJ))=hyt,Ptj)

where j is dropped from the income variable as it is not sector specific. The assumption about technology is made primarily for convenience, and any bias it may present is discussed below. Also note that in our estimation we use total GDP per capita rather and GDP originating in each sector. This is done partly due to data constraints, and partly due to inconsistencies between our objective and national accounting procedures, as will be discussed in the next section.

In order to allow for the possibility of a non-constant income elasticity

of energy demand we assume demand is of the form ec* = Aptlyt1+ y ny' . A logarithmic transformation of this demand function yields

'nectjj = Ц ,. +0, .) +bx'nptj . +b2'nyt ,. +&3(lny, ,)2 <D

where the subscript i denotes a particular country in the panel data set that will

be estimated.6 The variable A has been replaced by the term (ajti + 0, y ), which represents a country-specific effect and a time-specific effect respectively. The

5. We have in mind here some underlying theoretical demand framework that would yield energy use as a function of wealth and energy price. Since the energy variable is an aggregate measure, the price vector includes prices of all fuels available to the end-user at each point in time. When available, we use composite prices in the empirical analysis so that varying degrees of fuel substitution within total energy use is implicit.

6. The equation (1) can be extended to allow for non-linearity in price. Doing so, however, does not alter the conclusions reached herein.

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

86 / The Energy Journal

time effect, in particular, is meant to capture technological innovations or other time-specific peculiarities that are not accurately modeled by price and income, but that impact energy use. Testing procedures, which are explained in detail below, indicate that the time effect can reasonably be omitted. As a result, the

variable Btj is dropped from the remainder of our analysis. Equation (1) should be viewed as the long-run relationship between

energy use and income. It yields a long-run income elasticity of

b2+2b,'nytj (2)

which declines as income rises, provided b2 > 0 and b3 < 0. Equation (1) should only be regarded as an approximation to the relationship in question. In particular, if b2 > 0 and b3 < 0, it implies that there is some level of income, which we will call the "zero-point" because it is where the income elasticity of energy demand equals zero, at which per capita energy demand begins to decline with income. It is more likely, however, that per capita energy demand growth stagnates in relation to per capita GDP growth.

In order to capture short-run dynamics, we incorporate an adjustment

/ ' ( * V mechanism of the form f_ffL|=L^L / ' , which transforms to 'nect - 'nectA =

y('nec* - lnecM) upon taking logs.7 Adding the appropriate subscript, /, denoting

each individual country, the long-run model (1) becomes:

'nectjj = aji^ßl'nptji+ß2'nyti+ß3('nyti)2+(i-y)'nect_iji+8( (3>

where the aJti is a sector-specific individual country effect that can be treated as fixed or random. We estimate (3), noting that the long-run coefficients can be retrieved from bk = ßk ¡y for к = 1,2,3.

Equation (3) can be equivalently written in the form oo

'песф = BE(1 -y)kXt_kJ>i + со, *=o

where В is the vector of coefficients and X is the matrix of regressors. The coefficient у can thus be interpreted as an adjustment factor with a value between 0 and 1 . If у = 1 , then the adjustment to the long-run level of energy demand is instantaneous, and energy demand only depends on contemporaneous values of income and price. If у = 0, however, the adjustment process is infinitely long, in which case there is no geometric distribution of income and

7. We chose the Koyck (1954) partial adjustment mechanism rather than the error correction mechanism such as that used by Galli (1998) primarily because it preserves the variables in level form (i.e., it does not introduce any first-differenced variables into the equation (3)) while rendering

both short and long run elasticities.

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

Economie Development and End-Use Energy / 87

price effects over time so that energy demand today is affected equally by all past prices and incomes.

V. DATA

The data set used in our analysis consists of 28 countries,8 and covers the period from 1978 to 1995. The data were collected from a variety of sources (including individual country statistical yearbooks and international publications from the OECD, Asian Development Bank, the United Nations and the IEA), and were checked for both continuity and consistency. We use final energy consumption in each end-use sector (transportation, residential and commercial, and industrial and other) as our measure of energy demand. Accordingly, primary energy requirement is the sum of final demands in each sector plus transmission and conversion loss. The units of measurement are kilograms of oil

equivalent per capita. Therefore, being a physical quantity, the data are comparable across countries.

A purchasing power parity measure of per capita GDP, obtained from the Penn World Tables 5.6 (PWT), is used as the measure of income, and is denominated in 1985 international dollars. This measure allows for comparisons across countries because it accounts for differences in the purchasing power of different currencies. Real per capita growth rates were applied to the final observation in the PWT data set in order to obtain values up to 1995. We do not use a measure of GDP originating in each sector for three main reasons. One, purchaser exchange rates may be different across sectors. Applying a simple percentage to a PPP measure of GDP, therefore, could lead to erroneous estimates of the GDP originating in each sector. Two, national accounts are a measure of the value added within each sector. The value added in the

transportation sector, for example, is a measure of public and private transportation enterprises, but does not include the consumer's benefit of driving oneself. Since per capita income is a measure of consumer wealth, it will have better explanatory power of end-use energy demand. Three, the simple fact that we are seeking a relationship between the level of development and energy use in each sector warrants the use of per capita GDP as the explanatory variable.

Price data were the most elusive of the variables to collect. The energy

price index used for the transportation sector is an index of the real price of gasoline, and was available for most countries from the IEA with a few coming from national sources. While this is a less than perfect measure, neglecting for example differing degrees of dieselization, the prices for petroleum products should generally move together, reflecting the cost of crude oil.9 The energy

8. A complete list is given in Appendix A. 9. Unless there are changes in the relative rates at which each fuel is taxed.

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

88 / The Energy Journal

price index for the residential and commercial sector is the real price of fuel and light, and reflects the delivered cost of both electricity and heating oil. These data were obtained from both UN and national statistical sources. The indices

used for the transportation and residential and commercial sectors are based upon consumer prices, and therefore include any taxes on those particular fuels. The energy price index used for the industrial and other sector is the real producer or wholesale price of energy. This index was available for most countries from the IEA, but some data were obtained from national sources. Although these price indexes are less than perfect, they do capture information that both consumers and producers utilize in making decisions. The number of countries for which we could collect the relevant data limited our sample size.

VI. ESTIMATION PROCEDURE AND RESULTS

Standard estimation procedures (such as ordinary least squares) applied to the equation (3) may be inconsistent due to correlation between the lagged endogenous variable and the error term.10 Therefore, a suitable alternative must be chosen. We use the two-stage least squares (2SLS) approach based upon the work of Balestra and Nerlove (1966), using as instruments present and lagged values of the regressors and population. In choosing the set of instrumental variables, the treatment of the income and price variables as exogenous is crucial. If we assume that there exists an aggregate production function into which energy is an input, then the exogeneity assumption would certainly be incorrect.

The existence of a causal relationship between energy consumption and income has been investigated using a simple bivariate time-series model, but the results are mixed (see Kraft and Kraft, 1978; and Yu and Hwang, 1984, for competing results). However, bivariate tests may suffer from a bias due to omitted variables because certain variable relationships that may be important will not be included. Darrat, Gilley and Meyer (1996) use a vector autoregression to investigate the issue, thereby potentially avoiding the complication of omitted effects. They find evidence of unidirectional causality from GDP to energy consumption, suggesting that the growth of energy demand and GDP is not simultaneous (i.e., there is no evidence of bi-directional causality). However, Stern (1993, 2000), using a mulitvariate approach with a

10. See Maty as and Sevestre (1992), among others, for a complete discussion of this issue. Much theoretical work, notably that of Balestra and Nerlove (1966), Hsaio (1986), Arellano and Bond (1991) and Ahn and Schmidt (1995), has been done in the field of estimating dynamic panel data models. The more recent studies have been aimed at developing a 'better' estimator, on the grounds of increased efficiency. One applied study of considerable interest is that of Baltagi and Griffin (1997). They evaluate the out-of-sample forecast performance of a number of estimators that have been proposed for a dynamic panel data model applied to a very well researched field, gasoline demand.

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

Economie Development and End-Use Energy / 89

quality adjusted index measure of energy,11 found evidence of a causal relationship from energy to GDP in the US economy. Some authors have employed more rigorous modeling exercises in an effort to account for simultaneity. For example, Moroney (1992) attempts to identify the energy-GDP relationship by constructing a system of simultaneous equations based upon an assumed functional form for production technology. He shows that per capita energy demand is significant in explaining real per capita output, but reports parameter estimates for the income effect on energy that do not differ significantly from those found in studies that are less ambitious.

Although the evidence is inconclusive, we choose to proceed with treating income as exogenous. This choice is supported by the fact that we are using a sector-by-sector approach to modeling energy demand, which should alleviate any income-energy simultaneity that might exist. For example, it is unlikely that energy consumed in a particular sector, say transport, will have a significant effect on aggregate GDP.

The treatment of the price variable as exogenous is standard. It is based upon the notion that price is determined in an international market, and energy consumers are price takers in that market. Deviations from the international price are the result of different tax and fuel choice policies. These are assumed to be independent of the level of output and current energy consumption.

In determining the appropriate econometric specification, we first estimate the pooled model with constant intercept and slope (pooled 2SLS). This is then compared to the specification in which there are different intercepts with common slopes (2SLS-Within). The principal advantage of this estimator over its pooled 2SLS counterpart is that it allows for individual country effects.12 Hence, country specific heterogeneity can be explicitly taken into account. An F-test rejects the null of no individual country effects. The test is distributed as F(27,443) under the null of homogeneity, and yields a value of 3.46 in the industrial and other sector, 7.11 in the transport sector and 2.19 in the residential and commercial sector. All of these are significant at the 5% level.

Having established that there are different intercepts does not preclude the possibility that the slope parameters are also different. Pesaran and Smith (1995) claim that imposing homogeneity among the slope parameters across different cross-sectional units is overly restrictive and could result in biased

11. This amounts to using a composition weighted measure of energy, which is argued for on the grounds that the substitution of high quality energy sources such as electricity for lower quality energy sources such as coal can cause energy growth to appear distorted.

12. The presence of time effects was also tested using an F-test on the model with fixed and time effects versus the model with only fixed effects. Distributed as F(16,429) with critical value of 2.07 at the 5% level, these tests revealed that the null of no time effects cannot be rejected for the residential and commercial (F= 1 .69) and transport (F= 1 .39) sectors, but the null is rejected in the industrial sector (F=2.79). However, testing also reveals that the industrial data is not poolable (see below). For the industrial sector, therefore, this brings into question the power of the test for time effects. We therefore proceed with the fixed effect model for all sectors.

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

90 / The Energy Journal

estimates. To avoid this, they argue in favor of averaging the parameter estimates of the individual units, a method that produces consistent estimates as

long as N (number of cross-sectional units) and T (number of time periods) approach infinity. While their criticism has its merit, it is often ignored in many panel studies. Following Baltagi and Griffin (1997), we use the test based on the work of Wooldridge (1990) to ascertain whether the data can indeed be pooled. The numerator of this F-test is based upon the second stage residual sum of squares of 2SLS allowing for varying intercepts and slopes in its unrestricted version and varying intercepts and common slopes under its restricted version. The denominator is based on the unrestricted 2SLS residual sums of squares. The test is distributed as F(108,336) under the null of homogeneity, and yields a value of 2.47 for the industrial and other sector, 0.63 for residential and commercial, and 1.40 for the transportation sector.

The hypothesis of poolability is not rejected in the transport sector or the residential and commercial sector, but it is rejected in the industrial and other sector.13 This result is not very surprising. Differences in factor endowments and comparative advantages in production will be more pronounced in the industrial sector than in other sectors. In addition, government policy can distort the relative size and growth of the industrial sector from what comparative advantage would otherwise dictate. In some countries, industrial activity is highly protected through the use of quotas or tariffs, often motivated by the desire to increase domestic employment. Price controls are also often implemented in support of domestic interests. These policies have contributed to inefficient energy consumption practices. As such policies change, through liberalization measures for example, cost minimization efforts on the part of managers will increase efficiency, which will, in turn, impact industrial energy use. The resulting variability that is inherently present in the industrial sector not only affects the intercept term in (3), but also affects the slope coefficients across countries.

We opt to pool the data for all three sectors, despite the evidence in the industrial sector, on the grounds that the information obtained by doing so will provide insight into the global average pattern of energy demand growth. In deciding to pool the data, we note the finding of Baltagi and Griffin (1997), "Using a root mean square criterion, the efficiency gains from pooling appear to more than offset the biases due to inter-country heterogeneities" (p. 317). While the universality of this statement in its applicability to other data sets is certainly not proven, it does bring to light the possibility that pooled estimators in small samples contain more useful information than their single country counterparts, especially when making long term inferences.

13. Interestingly, the use of a log-linear specification yields a rejection of the null of poolability for all three sectors.

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

Economie Development and End-Use Energy / 91

This leaves us to determine the nature of the individual effect- i.e., is it fixed or random? The distinction between fixed and random effects is whether

or not the effect is correlated with the regressor set. If the effects are correlated with the other regressors, then a fixed effect specification is warranted. The fixed effect, or within, estimator, is always consistent. The gain from choosing the random specification where indicated is one of efficiency. A Hausman-type specification test, distributed x2 (4), indicates that the fixed effect approach is suitable for each sector. The test yields a value of 150.23 for the industrial and other sector, 14.08 for the residential and commercial sector, and 32.45 for the transportation sector. All are significant at the 5% level.

In Table 2, we report the parameter estimates, with standard errors in parentheses, for each end-use sector.14 In every case, we obtain plausible signs on the estimated coefficients. We also report the long-run coefficients that are implied by the adjustment process. The parameter estimates on the income variables indicate, by equation (2), that the income elasticity of energy demand does indeed fall as income rises. Furthermore, since the estimated coefficients are different across sectors, the elasticity is also different across sectors for a given level of income. This is indicative of the different rates of growth of energy demand in each of the end-use sectors, which is itself indicative of the changing structure of consumption and production.

The level of GDP per capita at the point when the income elasticity becomes zero (zero-points) can be calculated from the parameter estimates in

Table 2. Using the formula yepeak = expl- M, we find that the zero-points occur

at per capita income levels of $33,943 for the residential and commercial sector, $15,777 for the industrial and other sector, and $532,943 for the transportation sector. These zero-points should only be interpreted, however, as asymptotes. Per capita GDP in the United States in 1995 was over $19,000. Thus, our results would indicate that the zero-point in the industrial sector has been reached in the US. Due to the lag adjustment and price changes, however, it is possible to see further increases in industrial energy demand. For the residential and commercial sector, and particularly the transport sector, these results indicate that energy consumption will continue to increase well into the future, barring, of course, any significant increases in efficiency.

14. The parameter estimates for alternative methods of estimation are reported in Appendix B.

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

92 / The Energy Journal

Table 2. Estimation Results

Estimated Relationship (equation (3)): 'nectji = aJfi + /5,lqp,¿,- + № yu + ßßnyti)2 + (1 - 7) In ect.w + et

Long Run Relationship (equation (1)): 'nectji = aJfi + Ъх'прф + &2lnyu + b3Qnyu)2

Estimated Residential and Transportation Industrial and Other Coefficient Commercial

<*i

j8, -0.0939 -0.0938 -0.0664 (0.0216) (0.0124) (0.0157)

ß2 0.4632 0.4958 0.9531 (0.2034) (0.1371) (0.2228)

ft -0.0222 -0.0188 -0.0493 (0.0114) (0.0078) (0.0126)

1 - 7 0.9292 0.8196 0.7541 (0.0318) (0.0337) (0.0326)

R2 0.9266 0.9584 0.8907

LR Coefficient

bx -1.3263 -0.5200 -0.2700 b2 6.5424 2.7483 3.8760 b3 -0.3136 -0.1042 -0.2005

In Figure 2, we have used the long-run coefficients from Table 2 to generate the path of per capita energy demand for each sector and in sum, for a hypothetical average country. In doing this exercise we have assumed a fixed price, and the average of the country effects.15 The results illustrated in Figure 2 are meant to be illustrative of long-run tendencies for sector-by-sector energy demand. The data for any one country will certainly vary, as there are country- specific effects. We can see that the industrial sector grows the most rapidly at the outset of economic growth. This is followed by increases in energy demand in the other sectors to the point where energy used in transportation surpasses the energy consumed in the industrial sector.

15. The average effect is -1.48 in the residential and commercial sector, -2.64 in the industrial sector, and -1.39 in the transport sector. Price is held fixed at an index value of 100 (where 1985 = 100). Apparent in the simulated data is the assumption that energy consumption is 0 when GDP per capita is 0.

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

Economie Development and End-Use Energy 1 93

I Ô

I

I ее

а

I В

I ее

U

к

|

i

I 1

1 |

!

S со

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

94 / The Energy Journal

The estimated parameters in equation (3) reflect the effects of structural change and technology acquisition during the course of development. For example, as LDCs become wealthier they will be able to afford technologies in use by more developed nations. The development path in Figure 2, therefore, implicitly incorporates that both economic structure and technological status are changing through the course of development

Using the simulated data that generate the paths in Figure 2, we calculate and report, in Table 3, the share of total energy in each sector for various levels of per capital income. The trends implied by our results indicate that the energy consumed in the transportation sector will eventually exceed one- half of total final energy requirements. Although this will not occur until per capita incomes are well beyond what is currently observable, the suggested growth in transport fuel demand will have a substantial impact on world energy markets. Even if we allow for different fuel mixes for the economy in Figure 2, the fact that the majority of transport fuels are petroleum based indicates that crude oil will take up an increasing share of total energy as development progresses.

Table 3. Implied End-Use Sector Share of Total Final Energy Consumption

Real GDP/capita Residential and Transportation Industrial and 1985 international $ Commercial Other

$2,000 17.2% 20.5% 62.3% $6,000 30.8% 24.1% 45.1% $10,000 35.4% 27.1% 37.5% $14,000 37.3% 29.9% 32.8% $18,000 37.9% 32.4% 29.6% $22,000 38.0% 34.8% 27.2% $26,000 37.6% 37.1% 25.3% $30,000 37.1% 39.2% 23.7%

What do our results imply about the energy intensity of a nation as it develops? The energy intensity of an economy, which can be expressed at each

time / as -1

within each sector. The energy intensity profile for the hypothetical country in Figure 2 is recoverable from the simulated data and is illustrated in Figure 3.

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

Economie Development and End-Use Energy 1 95

j I 'S

ë 1

i" §

1

<N

§> Uh С

! i

'' § о eg

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

96 / The Energy Journal

The level of GDP per capita at the peak intensity of total final energy consumption is approximately $2,600, but the intensity of energy use profile is different across sectors. We can rewrite the above expression for energy

intensity as eCt>«"'Comm + ^hls. + ec'.w» . Each of the terms, therefore, represents yt yt У(

a contribution to the overall energy intensity at each time r,16 and each peaks at different levels of GDP per capita. In fact, these values can be calculated directly from the estimated coefficients shown in Table 2 using the formula ypeak

= exp - - for each sector. The industrial sector variable peaks at a per capita ' ~1Ьъ )

income of $1 ,322, the transport variable peaks at $4,395, and the residential and commercial variable peaks at $6,637. Since industrial consumption is the largest share of total energy use at low-income levels, overall energy intensity peaks at an income level close to the estimated industrial peak. However, increasing intensity of use in the non-industrial sectors contributes to a slow rate of decline in overall energy intensity after it peaks.

It is important to point out that we have modeled only final energy requirement. Primary energy requirement will be greater. The extent to which primary exceeds secondary requirement depends upon a number of factors. For example, heat yields of various grades of primary inputs are different. It may be possible, therefore, to reduce primary requirements while holding secondary use constant by substitution to fuels with a higher heat content. Another factor that influences the wedge between primary and secondary requirements is conversion loss. The use of primary resources for the generation of electricity or the refinement of petroleum products, for example, ultimately results in some conversion loss. The patterns above, therefore, must be coupled with some accounting of these factors on a country-by-country basis in order to assess the full impact of the growth of LDCs on the future of world energy markets.

VII. SAMPLE GROWTH PATTERNS

The trends laid out above are generalizations that capture average global tendencies. Developing countries can be expected to exhibit growth patterns similar to those illustrated in Figure 2, but there most certainly will be variation about those trends for any one country. However, the concerns about meeting energy resource requirements are shared by all economies because, as our analysis shows, energy demand will increase with the level of development. The extent to which energy demand is expected to increase in LDCs is of considerable importance. For example, if energy requirements in LDCs increase

16. Note that each term is not a sector-specific energy intensity.

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

Economie Development and End-Use Energy / 97

by enough, the burden to develop supplies sufficient to keep prices from substantially increasing could become overwhelming.

To illustrate this point, as an exercise, we have applied our parameter estimates to China, India, Indonesia and Brazil- the four largest developing nations in our sample- and the US in order to illustrate the pattern of growth implied by our results. The predicted trends illustrated in Figure 4 are the sum of the sectoral contributions within each country. The sectoral paths (not pictured) are corrected for individual country effects, and are generated to the year 2010 under the assumptions that the annual growth of per capita GDP is equal to the historical rate (1980 to 1995) and real prices remain constant at their 1995 levels.17 The population in each country is assumed to grow at the rate published in the World Bank Development Indicators, 1997.

An implication of declining energy intensity as the level of development increases is that for a given rate of growth, energy use will rise faster in less developed nations. Thus, as all nations become wealthier, we might expect some degree of convergence of the growth rate of per capita energy demand. This also implies that energy use in the more populous less developed nations, such as those depicted in Figure 4, will eventually surpass energy use in the US. In fact, at the projected rates of growth, Figure 4 indicates that total energy consumption in China will approach that in the US in the very near future. If China develops domestic resources to meet this need then coal use will rise considerably, which is undesirable for environmental reasons. If China turns outside its borders there

will be pressure internationally to develop non-OPEC energy supplies in order to prevent international reliance on OPEC from increasing. Significant increases in Chinese energy demand, and other LDCs for that matter, also has important implications for the implementation of the Kyoto Protocol. In particular, the fact that current negotiations do not require LDCs to take abatement measures means that the treaty will most likely do very little to reduce global carbon emissions.

What do our results imply about the composition of energy use? The share of energy consumed in the industrial and other sector will fall in each of these countries and the share of energy consumed in the residential and commercial sector and the transportation sector will rise. Moreover, our results suggest that energy demand in the transportation sector could increase by roughly a factor of three in China, more than double in Indonesia, almost double in India, and increase by about 25% in Brazil and the US. Energy demand in the residential and commercial sector exhibits similar behavior in each of the

countries. Given existing technologies, the predicted increase in demand in the transportation sector would naturally translate into an increasing dependence on oil and petroleum products. The extent to which this effect is either mitigated

17. The growth rates are 5. 1 % for China, 4.0% for India, 5.0% for Indonesia, 0.5 % for Brazil, and 1.4% for the US. Constant prices may not be reasonable, but is meant to be illustrative only. Precisely forecasting the energy demand of nations requires a more careful treatment of factors that determine energy use.

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

98 / The Energy Journal

i

I

г о

О n

s

I 1

!

(3D

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

Economie Development and End-Use Energy / 99

or exacerbated by fuel substitution in other sectors will determine how each of the developing giants adapts in future world oil markets. However, the degree of mitigation will most likely be minimal due to the fact that there is increased demand in those other sectors as well. In sum, the expectation that the developing giants are going to become increasingly important in world energy markets is most likely a well conceived one.

VIII. CONCLUDING REMARKS

In this paper we have examined the relationship between economic development and commercial energy demand. We have focused on the changes in the structure of production and consumption that accompany economic development and their effect on end-use energy demand in three end-use sectors: industrial and other, residential and commercial, and transportation. We have used the estimated long-run coefficients in our model to generate the path of per capita energy demand by sector and in total, for a hypothetical country, in order to demonstrate the effect that patterns of development have on end-use energy consumption.

Our interest has not been to develop a precise forecasting model but rather to understand the effect of economic development on the use of commercial energy. The path of energy use that we have constructed reflects the effects of structural change and technology adoption during the course of development, and is meant to convey patterns only. A forecasting model would have to take into account a number of variables, which are implicitly held constant in the construction of the energy use paths in Figure 4. For example, forecasts of real energy prices, real GDP growth rates, and population growth rates would be necessary to forecast demand. In addition, any forecasting exercise must also take into account the effects of energy and environmental policies that may be undertaken in the future.

As pointed out by Galli (1998), any valid exercise to model patterns of energy use as a function of economic development should allow for the energy intensity of a nation to rise then fall after some turning point. By using a log- quadratic specification, we are able to reproduce this result. In addition, by using a sectoral approach to modeling energy demand rather than one which models total demand directly, we are able to gauge the impact of development on each sector's contribution to total energy intensity. Since different sectors of an economy have different energy intensities and their share of total energy changes through the course of economic development (the shift from industry in Table 3, for example), total energy intensity is slow to fall, even as industrial energy intensity falls rapidly.

We have shown that transportation energy demand will eventually capture the largest portion of end-use energy consumption. Barring significant technological change, this translates into an increasing demand, and dependence, for crude oil, especially from developing nations. If this comes to pass, it would

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

100 / The Energy Journal

be reasonable to expect higher future oil prices. This could be due to things such as supply constraints in non-OPEC areas and/or OPEC exerting monopoly power over international oil markets as its share of world oil markets increases. This

outcome, however, may not occur for a number of inter-related reasons. In particular, rising energy prices will induce changes in technology and patterns of consumption that increase energy efficiency and reduce energy intensity. In addition, if the environment becomes of increasing concern, movements from dirty fuels to clean ones might occur. The effect on oil dependence is difficult to predict since the demand for oil will rise if producers substitute oil for coal (particularly in China and India). On the other hand, substitutions of gas for oil will have the opposite effect, although even these substitutions could be limited by price competitiveness. Finally, technological improvements in the exploration, development and production of oil (and gas) may result in lower (although, perhaps, volatile) prices.18

With increasing domestic need for energy resources, as well as international pressure to abate pollution, domestic policies will become increasingly important in determining future energy demand from conventional sources. Accordingly, in most countries there will be increasing pressures to institute policies to secure a stable flow of the 'cleaner' energy resources, such as oil and natural gas as opposed to coal.19 In the longer term, new technologies will replace current ones allowing for economic growth independent of growth in demand for fossil fuels, but the time horizon is uncertain. The need to meet

growing demands in the short term, therefore, will place pressures on governments to either develop domestic resource stocks or increase national dependence on international sources. What is certain is that energy will always be important for economic gain. Efficiency gains can reduce energy requirement per unit output, but not eliminate it.

Environmental concerns are capable of reducing particular types of energy use through both public awareness and international carbon tax policies. As implicit energy costs, such as those associated with pollution, increase or are internalized, substitution to cleaner fuels will be encouraged, perhaps through market intervention. The adoption of conservation policies and the encouragement of technological innovation in response to higher prices can both potentially have significant impacts on the future energy consumption patterns of all nations.

18. In a recent issue of Hart's "Oil and Gas World," Peter Aronstam, director of technology for

Baker Hughes, Inc. stated that the rate of technological progress in the oil and gas industry depends heavily upon the current economic environment, "In lean times... commercial introduction (of technologies) is delayed" (p. 17).

19. Coal (average US quality) emits approximately 26% more carbon per BTU than oil and 45 % more than natural gas.

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

Economie Development and End-Use Energy / 101

DATA SOURCES

Energy Prices and Taxes, International Energy Agency (IEA).

Energy Statistics Yearbook, various editions. United Nations.

Energy Balances of OECD and Non-OECD Countries, IEA.

OECD Historical Statistics, 1960-1995.

Penn World Tables 5.6 (http://cansim.epas.utoronto.ca).

Statistical Yearbooks for the various countries studied.

World Bank Development Indicators. World Bank.

APPENDICES

Appendix A

The countries, by geographic region, included in the study are as follows:

Asia/Pacific -

Pakistan, India, China, Indonesia, Thailand, Malaysia, South Korea, Japan, Australia.

Europe -

Turkey, Greece, Portugal, Spain, Ireland, Austria, Italy, Belgium, Netherlands, United Kingdom, France, Finland, Sweden, Denmark, Norway

North/South America -

Canada, United States, Mexico, Brazil

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

102 / The Energy Journal

Appendix В

The results from the alternative methods of estimation for each sector are as

follows:

Residential and Commercial -

OLS 2SLS

parameter std error parameter std error

constant -0.1395 0.2501 constant -0.1415 0.2502

у 0.1079 0.0603 у 0.1095 0.0605 у2 -0.0054 0.0036 у2 -0.0056 0.0037 р -0.0506 0.0138 р -0.0513 0.0139

ее, 0.9758 0.0043 ее, 0.9770 0.0052

Peak $21,823 Peak $17,620 GDP/cap GDP/cap

Within W-2SLS

parameter std error parameter std error

у 0.4991 0.1784 у 0.4632 0.2034 у2 -0.0240 0.0103 у2 -0.0222 0.0114 р -0.0965 0.0204 р -0.0939 0.0216

ее, 0.9204 0.0208 ее., 0.9292 0.0318

Peak $32,791 Peak $33,943 GDP/cap GDP/cap

Random-2SLS Average

parameter std error parameter

constant 0.5394 0.0286 constant 57.4225

у 0.0091 0.0070 у -11.4783 у2 0.0066 0.0004 у2 0.6423 р -0.1200 0.0011 р -0.1527

ее., 0.9022 0.0006 ее., 0.1971

Peak ... Peak

GDP/cap GDP/cap

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

Economie Development and End-Use Energy /103

Industrial and Other -

OLS 2SLS

parameter std error parameter std error

constant -0.1178 0.2805 constant -0.1408 0.2812

y 0.1124 0.0666 y 0.1188 0.0668 y2 -0.0078 0.0040 y2 -0.0085 0.0041 p -0.0494 0.0116 p -0.0510 0.0117

ec_, 0.9958 0.0065 ec_, 0.9999 0.0072

Peak $1,346 Peak $1,084 GDP/cap GDP/cap

Within W-2SLS

parameter std error parameter std error

y 0.7082 0.2002 y 0.9531 0.2228 y2 -0.0367 0.0115 y2 -0.0493 0.0126 p -0.0626 0.0156 p -0.0664 0.0157

ec, 0.8070 0.0251 ec_, 0.7541 0.0326

Peak $15,499 Peak $15,777 GDP/cap GDP/cap

Random-2SLS Average

parameter std error parameter

constant -0.6513 0.0202 constant 101.7700

y 0.2532 0.0048 y -20.7266 y2 -0.0155 0.0003 y2 1.0906 p -0.0612 0.0007 p -0.1522

ec_, 0.9875 0.0005 ec., 0.5676

Peak $3,525 Peak GDP/cap GDP/cap

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

104 / The Energy Journal

Transportation -

OLS 2SLS

parameter std error parameter std error

constant -0.5488 0.2608 constant -1.2956 0.3418

y 0.2132 0.0602 y 0.3895 0.0796 y2 -0.0098 О.ООЗЗ у2 -0.0150 0.0038 p -0.0605 0.0091 p -0.0638 0.0096

ec_, 0.9543 0.0091 ec_, 0.8888 0.0202

Peak $52,974 Peak $435,101 GDP/cap GDP/cap

Within W-2SLS

parameter std error parameter std error

y 0.5192 0.1332 y 0.4958 0.1371 y2 -0.0191 0.0078 y2 -0.0188 0.0078 p -0.0955 0.0121 p -0.0938 0.0124

ec_, 0.8020 0.0240 ec., 0.8196 0.0337

Peak $799,405 Peak $532,934 GDP/cap GDP/cap

Random-2SLS Average

parameter std error parameter

constant -5.0235 0.1314 constant 23.9931

y 1.4371 0.0314 y -4.8136 y2 -0.0260 0.0017 y2 0.3367 p -0.1101 0.0025 p -0.2399

ec, 0.1363 0.0070 ec., 0.0700

Peak $1 trillion Peak

GDP/cap GDP/cap

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms

Economie Development and End-Use Energy 1 105

REFERENCES

Ahn, S.C. and P. Schmidt (1995). "Efficient Estimation of Models for Dynamic Panel Data." Journal of Econometrics 68: 5-27.

Ang, B.W. (1987). "A Cross-Sectional Analysis of Energy-Output Correlation. " Energy Economics 9 (October): 274-286.

Arellano, M. and O. Bover (1995). "Another Look at Instrumental Variables Estimation of Error- Components Models." Journal of Econometrics 68: 29-51.

Balestra, Pietro and Marc Nerlove (1966). "Pooling Cross Section and Time Series Data in the Estimation of a Dynamic Model: The Demand for Natural Gas." Econometrica 34: 585-612.

Baltagi, Badi and James Griffin (1997). "Pooled Estimators vs. Their Heterogeneous Counterparts in the Context of Dynamic Demand for Gasoline." Journal of Econometrics 77: 303-327.

Bernardini, О. and R. Galli (1993). "Dematerialization: Long Term Trends in the Intensity of Use of Materials and Energy." Futures (May): 431-448.

Brookes, L.G. (1973). "More on the Output Elasticity of Energy Consumption." Journal of Industrial Economics (April): 83-94.

Chenery, Hollis B. and Moisés Syrquin (1975). Patterns of Development, 1950-1970. London: Oxford University Press.

Duey, Rhonda (1999). "Intelligent Completions," Hart Oil and Gas World, Phillips Business Information, Inc. (September): 17-18.

Galli, Rosanna (1998). "The Relationship Between Energy and Income Levels: Forecasting Long Term Energy Demand in Asian Emerging Countries." The Energy Journal 19(4): 85-105.

Hsiao, Cheng (1986). Analysis of Panel Data. Cambridge University Press. Judson, Ruth A. , Richard Schmalensee and Thomas M. Stoker (1999). "Economic Development and

the Structure of the Demand for Commercial Energy." The Energy Journal 20(2): 29-57.

Koyck, L.M. (1954). Distributed Lags and Investment Analysis. Amsterdam: North-Holland Publishing.

Kraft, J. and A. Kraft (1978). "On the Relationship Between Energy and GNP." The Journal of Energy and Development (Spring): 401-403.

Kusnets, Simon (1971). Economic Growth of Nations: Total Output and Production Structure. Cambridge, MA: The Belnapp Press of Harvard University Press.

Maty as Laszlo, and Patrick Sevestre (1992). The Econometrics of Panel Data: Handbook of Theory and Applications. Kluwer Academic Publishers.

Moroney, John R. (1992). "Energy, Capital, and Growth," in Advances in the Economics of Energy and Resources, Vol. 7.

Pesaran, M. H. and R. Smith (1995). "Estimating Long-Run Relationships from Dynamic Heterogeneous Panels." Journal of Econometrics 68: 79-113.

Pourgerami, Abbas and Christian R. von Hirschhausen (1991). Aggregate Demand tor bnergy and Dynamics of Energy Demand Elasticities in Non-Oil-Developing Countries." The Journal of Energy and Development 14(2): 237-251.

Presser, Richard D. (1985). "Demand Elasticities in the OECD." Energy Economics 7 (January): 9-12.

Stern, David (2000). "A Multivariate Cointegration Analysis of the Role of Energy in the US Macroeconomy." Energy Economics 22: 267-283.

Stern, David (1993). "Energy Use and Economic Growth in the USA, A Multivariate Approach." Energy Economics 15: 137-150.

Wooldridge, Jeffrey (1990). "A Note on the Lagrange Multiplier and F-statistics for Two Stage Least Squares Regressions." Economics Letters 34: 151-155.

Yu, Eden S.H. and Been-Kwei Hwang (1984). "The Relationship Between Energy and GDP." Energy Economics 6 (July): 186-190.

Zilberfarb, Ben-Zion and F. Gerard Adams (1981). "The Energy-GDP Relationship in Developing Countries." Energy Economics 3 (October): 244-248.

This content downloaded from 132.174.254.159 on Fri, 01 Nov 2019 18:11:52 UTC All use subject to https://about.jstor.org/terms