Content uploaded by Massimo Filippini

Author content

All content in this area was uploaded by Massimo Filippini

Content may be subject to copyright.

Energy demand and energy

efficiency in the OECD countries: a

stochastic demand frontier

approach

Massimo Filippini, Lester C. Hunt

CEPE Working Paper No. 68

Oktober 2009

CEPE

Zurichbergstrasse 18 (ZUE E)

CH-8032 Zurich

www.cepe.ethz.ch

Energy demand and energy efficiency in the OECD countries: a

stochastic demand frontier approach

Massimo Filippini and Lester C Hunt

Centre for Energy Policy and

Economics (cepe), ETH Zurich

and

Department of Economics,

University of Lugano,

Switzerland

Surrey Energy Economics

Centre (SEEC) and Research

Group on Lifestyles Values and

Environment (RESOLVE),

Department of Economics,

University of Surrey, UK

Abstract

This paper attempts to estimate a panel ‘frontier’ whole economy aggregate energy demand

function for 29 countries over the period 1978 to 2006 using stochastic frontier analysis (SFA).

Consequently, unlike standard energy demand econometric estimation, the energy efficiency of

each country is also modelled and it is argued that this represents a measure of the underlying

efficiency for each country over time, as well as the relative efficiency across the 29 OECD

countries. This shows that energy intensity is not necessarily a good indicator of energy

efficiency, whereas by controlling for a range of economic and other factors, the measure of

energy efficiency obtained via this approach is. This is, as far as is known, the first attempt to

model energy demand and efficiency in this way and it is arguably particularly relevant in a

world dominated by environmental concerns with the subsequent need to conserve energy

and/or use it as efficiently as possible. Moreover, the results show that although for a number

of countries the change in energy intensity over time might give a reasonable indication of

efficiency improvements; this is not always the case. Therefore, unless this analysis is

undertaken, it is not possible to know whether the energy intensity of a country is a good proxy

for energy efficiency or not. Hence, it is argued that this analysis should be undertaken to

avoid potentially misleading advice to policy makers.

JEL: D, D2, Q, Q4, Q5.

Keywords: Energy demand; OECD; efficiency and frontier analysis; energy efficiency.

Acknowledgements

We are grateful to Olutomi Adeyemi for his assistance with the data collection. A preliminary version of the paper

was presented at the 2nd International workshop on Empirical Methods in Energy Economics (Jasper, Canada,

2009) and we are grateful to the discussant, Denise Young and other participants for their very helpful comments

and suggestions. A revised version of the paper was presented at the 10th IAEE European conference (Vienna,

Austria, 2009) and we thank participants for their additional comments and suggestions. The authors are, of

course, responsible for all errors and omissions.

EnergydemandandenergyefficiencyintheOECDcountries:astochasticdemandfrontierapproachPage1of23

1 Introduction

During the last 20 years, there has been considerable debate within energy policy about

the possible contribution from an improvement in energy efficiency and on the effectiveness of

ecological tax reforms in the alleviation of the greenhouse effect and in the decrease of the

dependency on fossil fuels. In order to design and implement effective energy policy

instruments to promote an efficient and parsimonious utilization of energy, it is necessary to

have information on energy demand price and income elasticities in addition to sound

indicators of energy efficiency.

In practical energy policy analysis, the typical indicator used is energy intensity,

defined as the ratio of energy consumption to GDP. This is highlighted by a report from the

International Energy Agency (IEA, 2009) on the Energy Efficiency Policies in the G8, which

states that since the 1970s many countries have promoted energy efficiency improvements,

which is illustrated by the decline in energy intensity. The report goes on to say that “Energy

intensity is the amount of energy used per unit of activity. It is commonly calculated as the

ratio of energy use to GDP. Energy intensity is often taken as a proxy for energy efficiency,

although this is not entirely accurate since changes in energy intensity are a function of

changes in several factors including the structure of the economy and energy efficiency” (our

emphasis, p. 15). This highlights the weakness of this simple aggregate energy consumption to

GDP ratio in that it does not measure the level of ‘underlying energy efficiency’ that

characterizes an economy; hence, it is difficult to make conclusions for energy policy based

upon this simple measure.

In this paper, an alternative way to estimate the economy-wide level of energy

efficiency is proposed, by drawing on different strands of the energy economics research

EnergydemandandenergyefficiencyintheOECDcountries:astochasticdemandfrontierapproachPage2of23

literature; in particular, frontier estimation and energy demand modelling. An energy demand

frontier function is therefore estimated in order to attempt to isolate ‘underlying energy

efficiency’, by explicitly controlling for income and price effects, country specific effects,

climate effects and a common Underling Energy Demand Trend (the UEDT, capturing both

‘exogenous’ technical progress and other exogenous factors). Hence, it allows for the impact

of ‘endogenous’ technical progress’ through the price effect and ‘exogenous’ technical

progress through the UEDT.

The aim is to analyse economy wide energy efficiency; hence, the estimated model

introduced below is for aggregate energy consumption for the whole economy. Economy wide

aggregate energy demand is derived from the demand for energy services such as heat,

illumination, cooked food, hot water, transport services, manufacturing processes, etc. To

produce the desired services it is generally necessary to use a combination of energy fuels and

capital equipment such as household appliances, cars, insulated walls, machinery, etc. This

implies that the demand for energy is influenced by the level of energy efficiency of the

equipment and, generally, of the production process. For instance, some relatively new

equipment and production processes are able to provide the same level of services and products

using less energy than old equipment. This comes from research and development that

improves the thermodynamic efficiency of appliances and the capital stock, as well as

production processes – there is a technical improvement. Of course, in reality, apart from the

technological and economic factors there are a range of exogenous institutional and regulatory

factors that are important in explaining the level of energy consumption, furthermore, these

exogenous changes are unlikely to impact in a consistent rate over time. Hence, it is important

that the UEDT is specified in such a way that it is ‘non-linear’ and could increase and/or

EnergydemandandenergyefficiencyintheOECDcountries:astochasticdemandfrontierapproachPage3of23

decrease over the estimation period as advocated by Hunt et al. (2003a,b). Therefore, given a

panel data set is used this is achieved by time dummies as proposed by Griffin and Schulman

(2005) and Adeyemi and Hunt (2007).

In order to try to tease out these different influences, a general energy demand

relationship found in the standard energy demand modelling literature, relating energy

consumption to economic activity and the real energy price, is utilised for the estimation of an

aggregate energy demand function for a panel of OECD countries. Moreover, in order to

control for other important factors that vary across countries and hence can affect a country’s

energy demand, some variables related to climate, size, and structure of the economy are

introduced in the model. Thus the framework adopted here attempts to isolate the ‘underlying

energy efficiency’ for each country after controlling for income, price, climate effects,

technical progress and other exogenous factors, as well effects due to difference in area size

and in the structure of the economy. The estimated model therefore isolates the level of

underlying energy efficiency, defined with respect to a benchmark, e.g. a best practice

economy in the use of energy by estimation a ‘common energy demand’ function across

countries, with homogenous income and price elasticities, and responses to other factors, plus a

homogenous UEDT. This is seen as important, given the need to isolate the different

underlying energy efficiency across the countries.1 Consequently, once these effects are

adequately controlled for, it allows for the estimation of the underlying energy efficiency for

each country showing i) how efficiency has changed over the estimation period and ii) the

differences in efficiency across the panel of countries.

1 The UEDT includes exogenous technical progress and it could be argued that even though technologies are

available to each country they are not necessarily installed at the same rate; however, it is assumed that this results

from different behaviour across countries and reflects ‘inefficiency’ across countries; hence, it is captured by the

different (in)efficiency terms for all countries.

EnergydemandandenergyefficiencyintheOECDcountries:astochasticdemandfrontierapproachPage4of23

The paper is organized as follows. The next section, discusses the rationale and

specification of the energy demand frontier function, with the data and econometric

specification introduced in Section 3. The results of the estimation are presented in Section 4,

with a summary and conclusion in the final section.

2 An aggregate frontier energy demand model

Given the discussion above, it is assumed that there exists an aggregate energy demand

relationship for a panel of OECD countries, as follows:

Eit = E(Pit , Yit , Ci , Ai , ISHit , SSHit , Dt, EFit) (1)

where Eit is aggregate energy consumption per capita, Yit is GDP per capita, Pit is the real price

of energy, Ci is climate, Ai is the area size, ISHit is the share of value added of the industrial

sector and SSHit is the share of value added for the service sector all for country i in year t.

Further, Dt is a series of time dummy variables representing the UEDT that captures the

common impact of important unmeasured exogenous factors that influence all countries

simultaneously, e.g. general expectations of changes in international oil price, general changes

in awareness of climate change, and exogenous change in the technology. Finally, EFit is the

level of ‘underlying energy efficiency’ of the appliance and capital equipment used in an

economy. This could incorporate a number of factors that will differ across countries,

including different government regulations as well as different social behaviours, norms,

lifestyles and values. Hence, a low level of underlying energy efficiency implies an inefficient

use of energy (i.e. ‘waste energy’), so that in this situation, awareness for energy conservation

could be increased in order to reach the ‘optimal’ energy demand function. Nevertheless, from

an empirical perspective, when using OECD aggregate energy data, the aggregate level of

EnergydemandandenergyefficiencyintheOECDcountries:astochasticdemandfrontierapproachPage5of23

energy efficiency of the capital equipment and of the production processes is not observed

directly. Therefore, this underlying energy efficiency indicator has to be estimated.

Consequently, in order to estimate this economy-wide level of underlying energy efficiency

(EFit) and identify the best practice economy in term of energy utilization, the stochastic

frontier function approach introduced by Aigner et al. (1977) is used.2

The stochastic frontier function has generally been used in production theory to

measure, using an econometric approach, the economic performance of production processes.

The central concept of the frontier approach is that in general the function gives the maximum

or minimum level of an economic indicator attainable by an economic agent. For a production

function, the frontier gives the maximum level of output attainable by a firm for any given

level of inputs. In the case of an aggregate energy demand function, used here, the frontier

gives the minimum level of energy necessary for an economy to produce any given level of

energy services. In principle, the aim here is to apply the frontier function concept in order to

estimate the baseline energy demand, which is the frontier that reflects the demand of the

countries that use high efficient equipment and production process. This frontier approach

allows the possibility to identify if a country is, or is not, on the frontier. Moreover, if a country

is not on the frontier, the distance from the frontier measures the level of energy consumption

above the baseline demand, e.g. the level of energy inefficiency.

The approach used in this study is therefore based on the assumption that the level of

the economy-wide energy efficiency can be approximated by a one-sided non-negative term, so

2 Of course, the frontier function approach suggested by Aigner et al. (1977) has been developed within the

neoclassical production theory. The main goal of this literature has been to estimate production and cost frontier in

order to identify the level of productive inefficiency (allocative and technical inefficiency). In this study, the

neoclassical production theory is discarded and instead the concept of a stochastic frontier within the empirical

approach traditionally used in the estimation of economy-wide energy demand function is employed. Of course,

behind the concept of underlying energy inefficiency developed here, there is still a ‘production process’.

EnergydemandandenergyefficiencyintheOECDcountries:astochasticdemandfrontierapproachPage6of23

that a panel log-log functional form of Equation (1) adopting the stochastic frontier function

approach proposed by Aigner et al. (1977) can be specified as follows:

(2)

ititit

S

it

I

i

a

i

C

ttit

p

it

y

it uvSSHISHaDCDpye ++++++++=

ααααδααα

where eit is the natural logarithm of aggregate energy consumption per capita (Eit), yit is the

natural logarithm of GDP per capita (Yit), pit is the natural logarithm of the real price of energy

(Pit), DCi is a cold climate dummy variable, ai is the natural logarithm of the area size of a

country measured in squared km (Ai), ISHit is the share of value added of the industrial sector,

SSHit is the share of value added for the service sector and Dt is a series of time dummy

variables. Furthermore, the error term in Equation (2) is composed of two independent parts.

The first part, vit, is a symmetric disturbance capturing the effect of noise and as usual is

assumed to be normally distributed. The second part, uit, which represents the underlying

energy level of efficiency EFit in equation (1) is interpreted as an indicator of the inefficient

use of energy, e.g. the ‘waste energy’. It is a one-sided non-negative random disturbance term

that can vary over time, assumed to follow a half-normal distribution.3 An improvement in

the energy efficiency of the equipment or on the use of energy through a new production

process will increase the level of energy efficiency of a country. The impact of technological,

organisational, and social innovation in the production and consumption of energy services on

the energy demand is therefore captured in several ways: the time dummy variables, the

indicator of energy efficiency and through the price effect.4

3 It could be argued that this is a strong assumption for EF, but it does allow the ‘identification’ of the efficiency

for each country separately.

4 In this model specification, we are assuming that the price effect is symmetric. Gately and Huntington (2002),

amongst others, discuss the possibility of specifying a demand model with asymmetric price effects and some

EnergydemandandenergyefficiencyintheOECDcountries:astochasticdemandfrontierapproachPage7of23

In summary, Equation (2) is estimated in order to estimate underlying energy efficiency

for each country in the sample. The data and the econometric specification of the estimated

equations are discussed in the next section.

3. Data and econometric specification

The study is based on an unbalanced panel data set for a sample of 29 OECD countries

(i = 1, …, 29)5 over the period 1978 to 2006 (t = 1978-2006). This data set is based on

information taken from the International Energy Agency (IEA) database “World Energy

Statistics and Balances of OECD Countries” available at www.iea.org and from the general

OECD database “Country profile Statistics”.

E is each country’s per capita aggregate energy consumption in tonnes of oil equivalent

(toe), Y is each country’s per capita GDP in thousand US2000$PPP, and P is each country’s

index of real energy prices (2000=100). The climate dummy variable, DC, indicates whether a

country belongs to those characterized by a cold climate (according to the Köppen-Geiger

climate classification6) and A is the area size of a country is measured in squared kilometres.

Finally, the value added of the industrial and service sectors is measured as percentage of GDP

(ISH and SSH). Descriptive statistics of the key variables are presented in Table 1.

experimentation with asymmetric prices was undertaken here, however, the model did not fit the data well. Future

research will investigate this further.

5 Australia, Austria, Belgium, Canada, Czech Republic, Denmark, Finland, France, Germany, Greece, Hungary,

Ireland, Italy, Japan, Korea, Luxembourg, Mexico, Netherlands, New Zealand, Norway, Poland, Portugal, Slovak

Republic, Spain, Sweden, Switzerland, Turkey, the UK, and the US. For some countries, information on the share

of the industrial and service sector in the economy are only available for the years after 1990. For this reason the

data set is unbalanced.

6 See for a discussion of this classification Peel et al. (2007).

EnergydemandandenergyefficiencyintheOECDcountries:astochasticdemandfrontierapproachPage8of23

Table 1: Descriptive statistics

Variable Mean Std. Dev. Minimum Maximum

Description Name

Energy consumption per capita (toe/capita) E2.99 1.58 0.58 9.49

GDP per capita (1000 US2000$PPP/capita) Y20.63 8.44 4.19 63.36

Real Price of energy (2000=100) P99.65 16.42 53.56 170.30

Area size in km

2

A

1269850 2786260 2590 9984670

Share of industrial sector in % of GDP ISH 25.22 4.99 9.40 40.40

Share of service sector in % of GDP SSH 20.95 5.52 8.20 48.50

Climate Dummy DC 0.45 0.50 0 1

From the econometric specification perspective, the literature on the estimation of

stochastic frontier models using panel data needs to be considered. The first use of panel data

in stochastic frontier models goes back to Pitt and Lee (1981) who interpreted the panel data

random effects as inefficiency rather than heterogeneity.7 A major shortcoming of these

models is that any unobserved, time-invariant, group-specific heterogeneity is considered as

inefficiency. In order to solve this problem using panel data, Greene (2005a and 2005b)

proposed to extend the SFA model in its original form (Aigner, et al., 1977) by adding a fixed

or random individual effect in the model.8 It should be noted that these models produce

efficiency estimates that do not include the persistent inefficiencies that might remain more or

less constant over time. To the extent that there are certain sources of energy efficiency that

result in time-invariant excess energy consumption, the estimates of these models provide

relatively high levels of energy efficiency. For this reason, this study uses the original approach

proposed by Aigner, et al. (1977) so that fixed or random individual effects proposed by

Greene (2005a and 2005b) are not included in the model. Of course, by not considering the

individual effects in the econometric specification, it could result in the so-called ‘unobserved

7 Schmidt and Sickles (1984) and Battese and Coelli (1992) presented variations of this model.

8 For a successful application of these models in network industries, see Farsi, et al. (2006) and Farsi, et al. (2005).

EnergydemandandenergyefficiencyintheOECDcountries:astochasticdemandfrontierapproachPage9of23

variables bias’; e.g. a situation where correlation between observables and unobservables could

bias some coefficients of the explanatory variables. However, by introducing several

explanatory variables such as the climate, the area size, and some variables on the structure of

the economy it is possible to reduce this problem. In fact, the estimated coefficients of the

demand frontier function presented in the next section are very similar to those obtained by

estimating equation (2) by using a random or a fixed effects approach. 9 The econometric

approach used in this paper therefore has the advantage that it includes in the inefficiency term

the persistent inefficiencies that might remain more or less constant over time as well the

inefficiencies that vary over time.

Table 2 provides a summary of the model specification and a description of the

stochastic terms included in the model.

Table 2: Econometric specification of the model employed

Model Random error

εit

Level of efficiency

uit

TRE (ML)

ititit uv

+

=

ε

),0(iid~ 2

uit Nu

σ

+

),0(iid~ 2

vit Nv

σ

)( itit

uE

ε

The country’s efficiency is estimated using the conditional mean of the efficiency term

[

ititit vuuE +

]

, proposed by Jondrow et al. (1982). The level of energy efficiency can be

expressed in the following way:

)

ˆ

exp( it

it

F

it

it u

E

E

EF −== (3)

where Eit is the observed energy consumption per capita and is the frontier or minimum

demand of the ith country in time t. An energy efficiency score of one indicates a country on

F

it

E

9 In a preliminary analysis, a version of equation (2) using the true random effects model was also estimated. As

expected, the obtained level of energy efficiency were very high (average level of efficiency higher than 90%).

EnergydemandandenergyefficiencyintheOECDcountries:astochasticdemandfrontierapproachPage10of23

the frontier (100% efficient), while non-frontier countries, e.g. countries characterized by a

level of energy efficiency lower than 100%, receive scores below one. This therefore gives the

measure of underlying energy efficiency estimated below.10

In summary, Equation (2) is estimated and Equation (3) used to estimate the

efficiency scores for each country for each year. The results from the estimation are given in

the next section.

4. Estimation results

The estimation results for frontier energy demand model, Equation (2), are given in

Table 3. This shows that the estimated coefficients and lambda have the expected signs and are

statistically significant.11

Table 3: Estimated coefficients (t-values in parentheses)

Constant -1.916

(-6.93)

α

y 0.900

(38.98)

α

p

-0.275

(-4.77)

α

C

0.227

(12.29)

α

a 0.021

(3.44)

α

I

0.017

(9.08)

α

s

0.029

(11.51)

Time dummies Yes

Lamda (λ) 2.762

(8.71)

10 This is in contrast to the alternative indicator of energy inefficiency given by the exponential of uit. In this case,

a value of 0.2 indicates a level of energy inefficiency of 20%.

11 Lambda (λ) gives information on the relative contribution of uit and vit on the decomposed error term εit and

shows that in this case, the one-sided error component is relatively large.

EnergydemandandenergyefficiencyintheOECDcountries:astochasticdemandfrontierapproachPage11of23

For the variables in logarithmic form, the estimated coefficients can be directly

interpreted as elasticities. The estimated income elasticity and the estimated own price

elasticity are about 0.9 and -0.3 respectively, both not out of line with previous estimates. The

estimated area elasticity is about 0.02 indicating that a 10% larger country will demand 0.5%

more energy. The climate variable, DC, also appears to have an important influence on a

country’s energy demand; with countries characterized by a cold climate experiencing a higher

consumption of energy. Similarly, larger shares of a country’s industrial and service sectors

will also increase energy consumption. The time dummies, as a group, are significant and, as

expected, the overall the trend in their coefficients is negative as shown in Figure 1; however,

they do not fall continually over the estimation period, reflecting the ‘non-linear’ impact of

technical progress and other exogenous variables.

EnergydemandandenergyefficiencyintheOECDcountries:astochasticdemandfrontierapproachPage12of23

Table 4: Energy efficiency scores

min 0.522

max 0.951

mean 0.781

median 0.797

st.dev. 0.117

Table 4 provides descriptive statistics for the overall underlying energy efficiency

estimates of the countries obtained from the econometric estimation, showing that the mean

average efficiency is estimated to be about 78% (median 80%) nonetheless, as expected, there

is a fair degree of variation around the average. Table 5 presents the average energy efficiency

score for every country for three sub periods of the estimation period considered in the analysis

and over the whole period and Figure 2 shows that the estimated underlying energy efficiency

scores for each country over the estimation period relative to energy intensity. It should be

noted that, although presented individually for each country, the estimated efficiencies of each

country should not be taken as the precise position of each country given the stochastic

technique used in estimation. However, they do give a good relative indication of a country’s

change in efficiency over time and a country’s relative position vis-à-vis other countries.

Bearing this in mind, Table 5 and Figure 2 show that the estimated underlying energy

efficiency generally increased over the estimation period for some countries, such as Australia,

Canada, Denmark, Germany, Luxembourg, Netherlands, Norway, Sweden, the UK, and the

USA. Whereas for some countries the opposite is the case, with the estimated underlying

energy efficiency generally decreasing, such as Greece, Italy, Mexico, New Zealand, Portugal,

Spain and Turkey. Figure 2 also illustrates that the estimated underlying energy efficiency

would appear to be negatively correlated with energy intensity for most countries (i.e. the level

of energy intensity decreases with an increase of the level of energy efficiency), but with some

EnergydemandandenergyefficiencyintheOECDcountries:astochasticdemandfrontierapproachPage13of23

exceptions (discussed further below). This is to be expected in one sense. However, if this

technique were to be a useful tool for teasing out underlying energy efficiency then a perfect,

or even near perfect, negative correlation would not be expected since all the useful

information would be contained in the standard energy to GDP ratio.

Table 5: Average energy efficiency scores over time

1978 –

1987 1988 –

1997 1998 –

2006 Whole

Period

Australia0.768 0.783 0.806 0.785

Austria0.865 0.894 0.888 0.882

Belgium0.666 0.682 0.622 0.658

Canada0.583 0.608 0.645 0.608

CzechRepn/a 0.678 0.695 0.687

Denmark0.849 0.909 0.916 0.891

Finland0.581 0.584 0.612 0.591

France0.856 0.888 0.876 0.873

Germany0.844 0.931 0.944 0.905

Greece0.911 0.838 0.755 0.838

Hungaryn/a 0.742 0.823 0.788

Ireland0.628 0.725 0.902 0.747

Italy0.937 0.931 0.908 0.926

Japan0.880 0.890 0.863 0.878

Korea0.820 0.833 0.753 0.804

Luxembourg0.561 0.632 0.719 0.635

Mexico0.902 0.902 0.869 0.892

Netherlands0.612 0.681 0.701 0.663

NewZealand0.740 0.706 0.652 0.707

Norway0.790 0.802 0.864 0.817

Polandn/a 0.571 0.740 0.673

Portugal0.882 0.813 0.696 0.800

SlovakRep.n/a 0.594 0.637 0.622

Spain0.934 0.871 0.770 0.861

Sweden0.723 0.774 0.813 0.768

Switzerlandn/a 0.931 0.933 0.932

Turkey0.880 0.800 0.718 0.802

UK0.842 0.859 0.893 0.864

USA0.545 0.642 0.720 0.633

Note: n/a represents the situation where the average is not available over

the sub-period.

Due to the unbalanced panel, some averages are calculated over a

slightly shorter period than indicated.

EnergydemandandenergyefficiencyintheOECDcountries:astochasticdemandfrontierapproachPage14of23

EnergydemandandenergyefficiencyintheOECDcountries:astochasticdemandfrontierapproachPage15of23

EnergydemandandenergyefficiencyintheOECDcountries:astochasticdemandfrontierapproachPage16of23

This is confirmed, given the average correlation coefficient between the estimated

underlying energy efficiency and energy intensity across all countries is -0.68. Within this,

there is a relatively high negative correlation for some countries, such as Australia, Austria,

Canada, Denmark, Germany, Hungary, Ireland, Luxembourg, Netherlands, Norway, Poland,

Portugal, the Slovak Republic, Sweden, the UK and the USA; whereas for some countries the

(negative) correlation is somewhat less, such as Belgium, the Czech Republic, Greece, Japan,

Korea, New Zealand, and Switzerland. Furthermore, for Italy, Mexico, and Turkey, there

appears to be a positive relationship between the energy to GDP ratio and estimated energy

efficiency. This suggests that for some countries energy intensity is a reasonable proxy for

energy efficiency, whereas for others it is a very poor proxy. Hence, unless the analysis

undertaken here is conducted it is arguably not possible to identify for which countries energy

intensity is a good proxy and for which it is a poor proxy.

Turning to the differences in estimated energy efficiency scores across the panel of

countries in the sample it can be seen from Table 5 that there is some difference over the whole

sample period. Finland, Canada, the Slovak Republic, the USA, and Luxembourg are the

estimated five least efficient countries, with Switzerland, Italy, Germany, Mexico, and

Denmark the estimated five most efficient countries.12 This is further shown in Figure 3, with

the countries re-ordered from the most efficient to the least efficient. However, although Italy

is estimated to be one of the most energy efficient countries over time its level of efficiency has

been generally declining, despite a general fall in energy intensity. This highlights that energy

intensity in this case gives a poor indication of Italy’s change in energy efficiency over time.

12 However, it should be noted that, given the unbalanced panel used in estimation, the figures for the Slovak

Republic and Switzerland are over a much shorter period.

EnergydemandandenergyefficiencyintheOECDcountries:astochasticdemandfrontierapproachPage17of23

Countries will, however, have improved (or deteriorated) at different rates; hence,

Figure 4 gives the ordered data for the latter period only, 1998-2006. This shows that the

ordering does change, with the five least efficient countries being Finland, Belgium, the Slovak

Republic, Canada and New Zealand and the five most efficient countries being Germany,

Switzerland, Denmark, Italy and Ireland. Furthermore, as shown in Table 6, and illustrated

when comparing Figure 4 and Figure 5, it can be seen that although there is generally a

negative relationship between the rankings of the estimated underlying energy efficiency and

energy intensity there is not a one to one correspondence. For example, according to the

measure of energy intensity over the period 1998-2006, Germany is ranked 12th, whereas it is

estimated to be the most efficient over the period; suggesting that Germany is relatively more

energy efficient than the simple energy intensity measure would suggest. Conversely, Greece

and Portugal are ranked 1st and 12th respectively in terms of energy intensity but are only

ranked 16th and 23rd respectively in terms of underlying energy efficiency; suggesting that

EnergydemandandenergyefficiencyintheOECDcountries:astochasticdemandfrontierapproachPage18of23

Greece and Portugal are somewhat less energy efficient than the simple energy intensity

measure suggest.

EnergydemandandenergyefficiencyintheOECDcountries:astochasticdemandfrontierapproachPage19of23

Table 6: Comparison of the Rankings for Estimated Underlying Energy

Efficiency and Energy Intensity (1998-2006)

Estimated Underlying

Energy Efficiency

(symmetric model)

Energy Intensity (Energy

GDP ratio, toe per 1000

US2000$PPP)

Level Rank Level Rank

Australia0.806 14 0.130 17

Austria0.888 7 0.109 9

Belgium0.622 28 0.154 22

Canada0.645 26 0.213 29

CzechRep0.695 24 0.160 25

Denmark0.916 3 0.099 5

Finland0.612 29 0.184 28

France0.876 8 0.109 9

Germany0.944 1 0.114 12

Greece0.755 16 0.093 1

Hungary0.823 12 0.136 18

Ireland0.902 5 0.097 4

Italy0.908 4 0.093 1

Japan0.863 11 0.106 8

Korea0.753 17 0.160 25

Luxembourg0.719 20 0.156 24

Mexico0.869 9 0.112 11

Netherlands0.701 22 0.127 15

NewZealand0.652 25 0.152 21

Norway0.864 10 0.122 14

Poland0.740 18 0.142 20

Portugal0.696 23 0.114 12

SlovakRep.0.637 27 0.176 27

Spain0.770 15 0.103 7

Sweden0.813 13 0.140 19

Switzerland0.933 2 0.093 1

Turkey0.718 21 0.128 16

UK0.893 6 0.101 6

USA0.720 19 0.154 22

Note: A rank of 29 for underlying energy efficiency represents the least efficient

country by this measure, whereas a rank of 1 represents the most efficient

country. A rank of 29 for energy intensity represents the most energy intensity

country whereas a rank of 1 represents the least energy intensive country.

5. Summary and Conclusion

This research is a fresh attempt to isolate core energy efficiency for a panel of 29

OECD countries, opposed to relying on the simple energy to GDP ratio – or energy intensity.

EnergydemandandenergyefficiencyintheOECDcountries:astochasticdemandfrontierapproachPage20of23

By combining the approaches taken in energy demand modelling and frontier analysis, a

measure of the ‘underlying energy efficiency’ for each country is estimated. This approach has

not, as far is known, been attempted before. The energy demand specification controls for

income, price, climate country specific effects, area, industrial structure, and a underlying

energy demand trend in order to obtain a measure of ‘efficiency’ – in a similar way to previous

work on cost and production estimation – thus giving a measure of underlying energy

efficiency (reflecting the relative inefficient use of energy, i.e. ‘waste energy’).

The estimates for the core energy efficiency using this approach show that although for

a number of countries the change in energy intensity might give a reasonable indication of

efficiency improvements; this is not always the case both over time and across countries - Italy

and Greece being prime examples. For Italy, energy intensity declines over the estimation

period suggesting an improvement in energy efficiency, whereas the estimated underlying

energy efficiency falls over the period.13 For Greece, energy intensity suggests that it is the

most efficient country over the latter period covered by the data, whereas the estimated

underlying energy efficiency suggests otherwise. Therefore, unless the analysis advocated here

is undertaken, it is not possible to know whether the energy intensity of a country is a good

proxy for energy efficiency or not. Hence, it is argued that this analysis should be undertaken

in order to give policy makers an additional indicator other than the rather naïve measure of

energy intensity in order to try to avoid potentially misleading policy conclusions.

13 Although it still remains relatively one of the most efficient countries.

EnergydemandandenergyefficiencyintheOECDcountries:astochasticdemandfrontierapproachPage21of23

References

Adeyemi, O. I. and L. C. Hunt (2007) ‘Modelling OECD industrial energy demand:

asymmetric price responses and energy-saving technical change’, Energy Economics, 29(4),

pp. 693–709.

Aigner, D. J., C. A. K. Lovell and P. Schmidt (1977) ‘Formulation and Estimation of

Stochastic Frontier Production Function Models’, Journal of Econometrics, 6(1), pp. 21-37

Battese, G. E. and T. Coelli (1992) ‘Frontier production functions, technical efficiency and

panel data: with application to paddy farmers in India’, Journal of Productivity Analysis, 3,

pp. 153–69.

Farsi, M., M. Filippini and W. Greene (2005) ‘Efficiency Measurement in Network Industries:

Application to the Swiss Railway Companies’, Journal of Regulatory Economics, 28(1), pp.

69-90

Farsi, M., M. Filippini and , M. Kuenzle (2006) ‘Cost Efficiency in Regional Bus Companies:

An Application of Alternative Stochastic Frontier Models’, Journal of Transport Economics

and Policy, 40(1), pp. 95-118.

Gately, D. and H. G. Huntington (2002) ‘The asymmetric effects of changes in price and

income on energy and oil demand’, The Energy Journal, 23(1), pp. 19–55.

Greene, W. (2005a) ‘Reconsidering Heterogeneity in Panel Data Estimators of the Stochastic

Frontier Model’, Journal of Econometrics, 126, pp. 269-303

Greene, W. H. (2005b) ‘Fixed and random effects in stochastic frontier models’, Journal of

Productivity Analysis, 23(1), pp. 7–32.

Griffin J. M. and C. T. Schulman (2005) ‘Price asymmetry in energy demand models: A proxy

for energy-saving technical change?’, The Energy Journal, 26(2), pp. 1-21.

Hunt, L. C., G. Judge and Y. Ninomiya (2003a) ‘Underlying trends and seasonality in UK

energy demand: a sectoral analysis’, Energy Economics, 25(1), pp. 93–118.

Hunt, L. C., G. Judge and Y. Ninomiya (2003b) ‘Modelling underlying energy demand trends’,

Chapter 9 in: Hunt, L. C. (Ed.), Energy in a Competitive Market: Essays in Honour of Colin

Robinson, Edward Elgar, Cheltenham, pp. 140–174.

IEA (2009) ‘Progress with implementing energy efficiency policies in the G8’, International

Energy Agency Paper,

http://www.iea.org/Textbase/publications/free_new_Desc.asp?PUBS_ID=2127

Jondrow, J., C. A. K. Lovell, I. S., Materov and P. Schmidt (1982) ‘On the Estimation of

Technical Efficiency in the Stochastic Frontier Production Function Model’, Journal of

Econometrics, 19(2/3), pp. 233-238

Peel, M. C., B. L. Finlayson and T. A. and B. L. McMahon (2007) ‘Updated world map of the

Köppen-Geiger climate classification’. Hydrol. Earth Syst. Sci., 11, pp. 1633-1644.

EnergydemandandenergyefficiencyintheOECDcountries:astochasticdemandfrontierapproachPage22of23

EnergydemandandenergyefficiencyintheOECDcountries:astochasticdemandfrontierapproachPage23of23

Pitt, M. and L. Lee (1981) ‘The measurement and sources of technical inefficiency in the

Indonesian weaving industry’, Journal of Development Economics, 9, pp. 43–64.

Schmidt, P. and R. E. Sickles (1984) Production frontiers and panel data’, Journal of Business

and Economic Statistics, 2, pp. 367–74.