the rebound effect...the rebound effect: an assessment of the evidence for economy-wide energy...

The Rebound Effect:an assessment of the evidence for economy-wide

energy savings from improved energy efficiency

October 2007

Th

e R

eb

ou

nd

Effe

ct:an

asse

ssmen

t of th

e e

vid

en

ce fo

r eco

no

my-w

ide e

nerg

y sa

vin

gs fro

m im

pro

ved

en

erg

y e

fficien

cy

UK

En

erg

y R

ese

arch

Cen

tre

UK ENERGY RESEARCH CENTRE

58 Prince’s Gate

Exhibition Road

London SW7 2PG

tel: +44 (0)20 7594 1574

email: [email protected]

www.ukerc.ac.uk


S4134_cover 12/10/07 13:57

The Rebound Effect: an assessment of theevidence for economy-wide energy savingsfrom improved energy efficiency

A report produced by the Sussex Energy Group forthe Technology and Policy Assessment function ofthe UK Energy Research Centre

Steve Sorrell

October 2007

ISBN 1-903144-0-35

S4134_cover 12/10/07 13:57

This report has been produced by the UK Energy Research Centre’s Technology and PolicyAssessment (TPA) function.

The TPA was set up to address key controversies in the energy field throughcomprehensive assessments of the current state of knowledge. It aims to provideauthoritative reports that set high standards for rigour and transparency, while explainingresults in a way that is both accessible to non-technical readers and useful topolicymakers.

This report summarises the main conclusions from the TPA’s assessment of evidence fora rebound effect from improved energy efficiency. The subject of this assessment waschosen after extensive consultation with energy sector stakeholders and upon therecommendation of the TPA Advisory Group, which is comprised of independent expertsfrom government, academia and the private sector. The assessment addresses thefollowing question:

What is the evidence that improvements in energy efficiency will lead toeconomy-wide reductions in energy consumption?

The Summary Report seeks to present the main conclusions of this assessment in arelatively non-technical manner. The results of the full assessment are contained in fivein-depth Technical Reports, as follows:

1. Evidence from evaluation studies

2. Evidence from econometric studies

3. Evidence from elasticity of substitution studies

4. Evidence from CGE modeling studies

5. Evidence from energy, productivity and economic growth studies

A shorter Supplementary Note provides a graphical analysis of rebound effects. All thesereports are available to download from the UKERC website at: www.ukerc.ac.uk

The assessment was led by the Sussex Energy Group (SEG) at the University of Sussex,with contributions from the Surrey Energy Economics Centre (SEEC) at the University ofSurrey, the Department of Economics at the University of Strathclyde and ImperialCollege. The assessment was overseen by a panel of experts and is extremely wideranging, reviewing more than 500 studies and reports from around the world.

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

i

Preface

S4134 12/10/07 13:56 Page i

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

ii

S4134 12/10/07 13:56 Page ii

The UK Energy Research Centre’s mission is to be the UK’s pre-eminent centre of researchand source of authoritative information and leadership on sustainable energy systems. Itundertakes world-class research addressing the whole-systems aspects of energy supplyand use while developing and maintaining the means to enable cohesive research in energy.UKERC is funded by the UK Research Councils.

AcknowledgementsThis report has been written by Steve Sorrell (Sussex Energy Group, University of Sussex),but it is based upon the work of the full Project Team, which also includes: JohnDimitropoulos (Sussex Energy Group); Lester Hunt and David Broadstock (Surrey EnergyEconomics Centre, University of Surrey); Grant Allan, Michelle Gilmartin, Peter McGregor,Kim Swales and Karen Turner (Fraser of Allander Institute and Department of Economics,University of Strathclyde); and Matt Sommerville and Dennis Anderson (ICEPT, ImperialCollege).

John Dimitropoulos played a central role throughout the project, making an invaluablecontribution to all aspects of the research. Dennis Anderson provided two insightful papersthat greatly improved our understanding of the economy-wide rebound effect. Theremaining members of the Project Team co-authored one of the five Technical Reports thatprovide the foundation for this report (see Annex 2). While aspects of all their work havebeen incorporated into this report, individual members of the Project Team are notresponsible for its contents.

We have benefited enormously during this project from our interactions with HarrySaunders (Decision Processes Inc). We are also grateful for the insightful commentsreceived from our three peer reviewers, namely Manuel Frondel (ZEW), Karsten Neuhoff(University of Cambridge) and Jake Chapman. Valuable comments have also been receivedfrom Nick Eyre (Energy Savings Trust), Terry Barker (4CMR), Serban Scrieciu (4CMR), BlakeAlcott, Len Brookes, John Feather and Gordon Mackerron, as well as from participants atthe 29th IAEE International Conference (2006).

We are extremely grateful for the guidance, comments, patience and moral support offeredby Jim Skea (UKERC Research Director), Rob Gross (Head of UKERC TPA function), DennisAnderson (Senior Advisor, UKERC TPA) and Phil Heptonstall (UKERC TPA). Thanks also toPhil for copy editing.

Useful comments and advice have also been received from our Advisory Group, namely: JimSkea (UKERC); Horace Herring (Open University); Hunter Danskin (DEFRA); Tina Dallman(DEFRA); Ken Double (Energy Saving Trust); Lester Hunt (Surrey Energy EconomicsCentre); and Paolo Agnolucci (Policy Studies Institute).

This project has tried to summarise the work of hundreds of economists and analysts, manyof whom understand these issues much better than ourselves. A particular debt is owed toLorna Greening and David Greene for their previous synthesis of empirical work in this area(Greening and Greene, 1998).

The above individuals represent a range of views about the size of the economy-widerebound effect and none of them are responsible for the content of this report.

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

iii

About UKERC

S4134 12/10/07 13:56 Page iii

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

iv

S4134 12/10/07 13:56 Page iv

Most governments are seeking ways to improve energy efficiency in pursuit of their energypolicy goals. The potential ‘energy savings’ from improved energy efficiency are commonlyestimated using basic physical principles and engineering models. However, the energysavings that are realised in practice generally fall short of these engineering estimates.One explanation is that improvements in energy efficiency encourage greater use of theservices (for example heat or mobility) which energy helps to provide. Behaviouralresponses such as these have come to be known as the energy efficiency “rebound effect”.While rebound effects vary widely in size, in some cases they may be sufficiently large tolead to an overall increase in energy consumption - an outcome that has been termed‘backfire’. There is some evidence to suggest that improvements in the energy efficiencyof certain ‘pervasive’ technologies such as steam engines and electric motors havecontributed to backfire in the past.

Rebound effects are very difficult to quantify, and their size and importance under differentcircumstances is hotly disputed. Also, rebound effects operate through a variety ofdifferent mechanisms and lack of clarity about these has led to persistent confusion. Ingeneral, rebound effects have been neglected when assessing the potential impact ofenergy efficiency policies. A key conclusion of this report is that rebound effects are ofsufficient importance to merit explicit treatment. Failure to take account of rebound effectscould contribute to shortfalls in the achievement of energy and climate policy goals.

This report analyses the nature, operation and importance of rebound effects and providesa comprehensive review of the available evidence on this topic, together with closelyrelated issues, such as the link between energy consumption and economic growth. Itassesses the strengths and weaknesses of the evidence base, clarifies the underlyingdisputes and highlights the implications for energy and climate policy. The key messageis that promoting energy efficiency remains an effective way of reducing energyconsumption and carbon emissions. But more explicit treatment of rebound effects isneeded to assess the contribution that energy efficiency can realistically make.

Defining the rebound effect

• Many energy efficiency improvements do not reduce energy consumption by theamount predicted by simple engineering models. Such improvements make energyservices cheaper, so consumption of those services increases. For example, since fuel-efficient vehicles make travel cheaper, consumers may choose to drive further and/ormore often, thereby offsetting some of the energy savings achieved. Similarly, if afactory uses energy more efficiently it becomes more profitable encouraging furtherinvestment and greater levels of output. This is termed the direct rebound effect.

• Even if consumption of energy services remains unchanged, there are reasons whyenergy savings across the economy may be less than simple calculations suggest. Forexample, drivers of fuel-efficient cars may spend the money saved buying petrol onother energy-intensive goods and services, such as an overseas flight. Similarly, any

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

v

Executive Summary

S4134 12/10/07 13:56 Page v

reductions in energy demand will translate into lower energy prices which encourageincreased energy consumption. These mechanisms are collectively known as indirectrebound effects. The sum of direct and indirect rebound effects represents theeconomy-wide rebound effect. Rebound effects are normally expressed as apercentage of the expected energy savings from an energy efficiency improvement,so a rebound effect of 20% means that only 80% of the expected energy savings areachieved.

• Disputes over the magnitude of rebound effects arise in part from lack of clarity aboutdefinitions. Energy efficiency can be measured in a variety of ways, for example usingphysical indicators (tonnes of coal per tonne of steel) or economic ones (energy perunit of output measured in £). Energy efficiency can also be measured at differentlevels, for example for an individual manufacturing process, a factory, a company, asector or even the economy as a whole. Estimates of the rebound effect thereforedepend upon the indicators chosen and the level of analysis.

• Improvements in energy efficiency are often associated with improvements in theproductivity of capital, labour and materials. More efficient use of these other inputswill tend to amplify the rebound effect.

Estimating the rebound effect

• The available evidence for all types of rebound effect is far from comprehensive. Theevidence is better for direct effects than for indirect effects, but even this focuses ona small number of consumer energy services, such as home heating and personaltransportation, within developed countries. Both direct and indirect effects appear tovary widely between different technologies, sectors and income groups and in mostcases they cannot be quantified with much confidence. However the evidence doesnot suggest that improvements in energy efficiency routinely lead to economy-wideincreases in energy consumption. At the same time the evidence suggests thateconomy-wide rebound effects will be at least 10% and often higher. Rebound effectstherefore need to be factored into policy assessments.

• For household heating, household cooling and personal automotive transport indeveloped countries, the direct rebound effect is likely to be less than 30% and maybe closer to 10% for transport. Direct rebound effects for these energy services arelikely to decline in the future as demand saturates. Improvements in energy efficiencyshould therefore achieve 70% or more of the reduction in energy consumptionprojected using engineering principles. However, indirect effects mean that theeconomy-wide reduction in energy consumption will be less.

• Direct rebound effects are likely to be smaller where energy forms a relatively smallproportion of total costs and has little influence on operating decisions. The use ofelectricity in electronic appliances is a good example. However, these effects have only

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

vi

S4134 12/10/07 13:56 Page vi

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

vii

been studied over relatively limited time periods. There are also some importantuncertainties, for example about the link between direct rebound effects andhousehold income.

• There are very few studies of rebound effects from energy efficiency improvements indeveloping countries. Rebound effects may be expected to be larger in developingcountries where demand for energy services is far from saturated. This is supportedby the limited empirical evidence available. In some cases, the direct rebound effectmay exceed unity.

• Energy-economic models can be used to estimate indirect and economy-wide reboundeffects, but there are few published studies and these have a number of flaws. Theresults demonstrate that the economy-wide rebound effect varies widely dependingupon the sector where the energy efficiency improvement takes place. While littleconfidence can be placed in the available estimates, several studies suggest thateconomy-wide rebound effects may frequently exceed 50% (i.e. less than half of theexpected energy savings will be achieved). Moreover, these estimates do not take intoaccount the amplifying effect of any associated improvements in the efficiency withwhich capital, labour or materials is used.

Rebound or backfire?

• The so-called ‘Khazzoom-Brookes (K-B) postulate’ claims that, if energy prices do notchange, cost effective energy efficiency improvements will inevitably increaseeconomy-wide energy consumption above what it would be without thoseimprovements (‘backfire’). This provocative claim would have serious implications forenergy and climate policy if it were correct. However, the theoretical arguments infavour of the postulate rely upon stylised models that have a number of limitations,such as the assumption that economic resources are allocated efficiently. Similarly, theempirical evidence for the postulate is indirect, suggestive and ambiguous. Since anumber of flaws have been found with both the theoretical and empirical evidence, theK-B ‘hypothesis’ cannot be considered to have been verified. Nevertheless, thearguments and evidence used to defend the postulate deserve more serious attentionthan they have received to date.

• In developed countries, energy use as conventionally measured has grown moreslowly than the economy as a whole. From this, it is generally concluded that technicalchange has improved the efficiency with which energy is used and thereby helped to‘decouple’ energy consumption from economic growth. However once different energysources are weighted by their relative ‘quality’ or economic productivity, the couplingbetween energy consumption and economic growth appears far stronger. Takentogether, the evidence reviewed in this report suggests that: a) the scope forsubstituting other inputs for energy is relatively limited; b) much technical change hashistorically increased energy intensity; c) energy may play a more important role in

S4134 12/10/07 13:56 Page vii

economic growth than is conventionally assumed; and d) economy-wide reboundeffects may be larger than is conventionally assumed.

• If improvements in energy efficiency are associated with improvements in theefficiency with which other inputs such as capital, labour and materials are used, thenit is quite possible that energy consumption may increase as the K-B postulatesuggests. However, since the link between the efficiency with which energy and otherinputs is used depends on individual technologies and circumstances, there is no apriori reason to believe that ‘backfire’ is an inevitable outcome in all cases.

• The debate over the K-B postulate would benefit from clearer distinctions betweendifferent types of energy efficiency improvement. For example, the K-B postulateseems more likely to hold for improvements associated with pervasive ‘general-purpose technologies’, particularly when these are adopted by producers rather thanfinal consumers and when the improvements occur at an early stage of developmentand diffusion. Steam engines provide a good illustration from the 19th-century, whileelectric motors provide a comparable illustration from the early 20th century. Theopportunities offered by these technologies have such significant effects oninnovation, productivity and economic growth that economy-wide energy consumptionis increased over the long-term. Such technologies are generally taken upenthusiastically by the market and are rarely associated with purposeful public policy.

• In contrast, the K-B postulate seems less likely to hold for dedicated energy efficiencytechnologies such as thermal insulation, particularly when these are used byconsumers or when they play a subsidiary role in economic production. Thesetechnologies have smaller effects on productivity and economic growth, with the resultthat economy-wide energy consumption is likely to be reduced.

Policy implications

1. The potential contribution of energy efficiency policies needs to bereappraised.

• Energy efficiency may be encouraged through policies that raise energy prices, suchas carbon taxes, or through non-price policies such as building regulations. Bothshould continue to play an important role in energy and climate policy. However, manyofficial and independent appraisals of such policies have undoubtedly overstated thecontribution of non-price policies to reducing energy consumption and carbonemissions.

• It would be wrong to assume that, in the absence of evidence, rebound effects are sosmall that they can be disregarded. Under some circumstances (e.g. energy efficienttechnologies that significantly improve the productivity of energy intensive industries)economy-wide rebound effects may exceed 50% and could potentially increase energyconsumption in the long-term. In other circumstances (e.g. energy efficiency

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

viii

S4134 12/10/07 13:56 Page viii

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

ix

improvements in consumer electronic goods) economy-wide rebound effects are likelyto be smaller. But in no circumstances are they likely to be zero.

• Taking rebound effects into account will reduce the apparent effectiveness of energyefficiency policies. However, many energy efficiency opportunities are highly cost-effective and will remain so even when rebound effects are allowed for. Providedmarket and organisational failures can be overcome, the encouragement of theseopportunities should increase real income and contribute to economic growth. Theymay not, however, reduce energy consumption and carbon emissions by as much aspreviously assumed.

2. Rebound effects should be taken into account when developing andtargeting energy efficiency policy

• Rebound effects vary widely between different technologies, sectors and incomegroups. While these differences cannot be quantified with much confidence, thereshould be scope for including estimated effects within policy appraisals and usingthese estimates to target policies more effectively. Where rebound effects areexpected to be large, there may be a greater need for policies that increase energyprices.

• ‘Win-win’ opportunities that reduce capital and labour costs as well as energy costsmay be associated with large rebound effects. Hence, the implications of encouragingthese opportunities need to be clearly understood and quantified. It may make moresense to focus policy on ‘dedicated’ energy efficient technologies, leaving therealisation of wider benefits to the market

3. Rebound effects may be mitigated through carbon/energy pricing –whether implemented through taxation or an emissions trading scheme

• Carbon/energy pricing can reduce direct and indirect rebound effects by ensuring thatthe cost of energy services remains relatively constant while energy efficiencyimproves. Carbon/energy pricing needs to increase over time at a rate sufficient toaccommodate both income growth and rebound effects, simply to prevent carbonemissions from increasing. It needs to increase more rapidly if emissions are to bereduced.

• Carbon/energy pricing may be insufficient on its own, since it will not overcome thenumerous barriers to the innovation and diffusion of low carbon technologies andcould have adverse impacts on income distribution and competitiveness. Similarly,policies to address market barriers may be insufficient, since rebound effects couldoffset much of the energy savings. A policy mix is required.

S4134 12/10/07 13:56 Page ix

AEEI Autonomous Energy Efficiency Improvements

AES Allen-Urzwa Elasticity of Substitution

CES Constant Elasticity of Substitution

CGE Computable General Equilibrium

CPE Cross Price Elasticity

DEFRA Department for the Environment, Food and Rural Affairs

EBPP Evidence Based Policy and Practice

GDP Gross Domestic Product

GPT General Purpose Technologies

GWh Gigawatt Hour

IAEE International Association for Energy Economics

IEA International Energy Agency

K-B Khazzoom-Brookes

kW Kilowatt - a measure of power, one thousand W

kWh Kilowatt hours - a measure of energy, one kW of power provided for one hour

MES Morishima Elasticity of Substitution

MRTS Marginal Rate of Technical Substitution

OECD Organisation for Economic Cooperation and Development

SUV Sport Utility Vehicle

TFP Total Factor Productivity

TPA UKERC Technology and Policy Assessment function

UKERC UK Energy Research Centre

ZEW Zentrum für Europäische Wirtschaftsforschung, Mannheim

4CMR Cambridge Centre for Climate Change Mitigation Research

Glossary

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

x

S4134 12/10/07 13:56 Page x

1. Introduction ....................................................................................... 11.1 Context and rationale .................................................................... 11.2 How the assessment was conducted ................................................ 61.3 Report structure............................................................................ 9

2. What is energy efficiency?.................................................................. 112.1 Energy efficiency and energy savings............................................... 112.2 Multiple definitions ........................................................................ 122.3 Unacknowledged implications.......................................................... 152.4 Summary ..................................................................................... 17

3. The evidence for direct rebound effects ............................................. 193.1 Understanding the direct rebound effect........................................... 193.2 Estimating direct rebound effects from evaluation studies................... 223.3 Estimating direct rebound effects from econometric studies ................ 263.4 Direct rebound effects for private automotive transport...................... 303.5 Direct rebound effects for other consumer energy services ................. 323.6 Sources of bias in estimates of the direct rebound effect .................... 373.7 Summary ..................................................................................... 38

4. The evidence for indirect and economy-wide rebound effects ............ 414.1 Embodied energy – the limits to substitution .................................... 414.2 Secondary effects.......................................................................... 434.3 Evidence for limits to substitution.................................................... 454.4 Evidence for secondary effects ........................................................ 484.5 Evidence for economy-wide effects from general equilibrium models .... 514.6 Evidence for economy-wide effects from macro-econometric models .... 574.7 Summary ..................................................................................... 59

5. The evidence for ‘backfire’.................................................................. 615.1 Historical perspectives ................................................................... 625.2 Energy productivity and economic growth......................................... 645.3 Energy productivity and production theory........................................ 715.4 Substitution between energy and capital .......................................... 735.5 Energy productivity and ecological economics ................................... 775.6 Implications.................................................................................. 835.7 Summary ..................................................................................... 84

6. Conclusions, research needs and policy implications.......................... 876.1 Conclusions .................................................................................. 876.2 Research needs............................................................................. 896.3 Policy implications ......................................................................... 92

References.................................................................................................. 95

Annex 1: Project team, expert group and contributors............................... 107

Annex 2: Technical and Supplementary Reports......................................... 108

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

xi

Contents

S4134 12/10/07 13:56 Page xi

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

xii

S4134 12/10/07 13:56 Page xii

The UKERC Technology and PolicyAssessment (TPA) function was set up toaddress key controversies in the energyfield through comprehensive assessmentsof the current state of knowledge. It aimsto provide authoritative reports that sethigh standards for rigour andtransparency, while explaining results in away that is both accessible to non-technical readers and useful topolicymakers. This latest report addressesthe following question:

What is the evidence thatimprovements in energy efficiencywill lead to economy-wide reductionsin energy consumption?

1.1 Context and rationaleTo achieve reductions in carbonemissions, most governments are seekingways to improve energy efficiencythroughout the economy. It is generallyassumed that such improvements willreduce overall energy consumption, atleast compared to a scenario in whichsuch improvements are not made. But arange of mechanisms, commonly groupedunder the heading of rebound effects mayreduce the size of the ‘energy savings’achieved. Indeed, there is some evidenceto suggest that the introduction of certaintypes of energy efficient technology in thepast has contributed to an overallincrease in energy demand – an outcomethat has been termed ‘backfire’. Thisappears to apply in particular to pervasivenew technologies, such as steam enginesin the 19th century, that raise overalleconomic productivity as well asimproving energy efficiency.

These rebound effects could have far-reaching implications for energy andclimate policy, both in the UK and globally.While cost-effective improvements inenergy efficiency should improve welfareand benefit the economy, they could insome cases provide an ineffective or evena counterproductive means of tacklingclimate change. However, it does notnecessarily follow that all improvementsin energy efficiency will increase overallenergy consumption or in particular thatthe improvements induced by policymeasures will do so.

The nature, operation and importance ofrebound effects are the focus of a long-running dispute within energy economics.On the micro level, the question iswhether improvements in the technicalefficiency of energy use can be expectedto reduce energy consumption by theamount predicted by simple engineeringcalculations. For example, will a 20%improvement in the fuel efficiency ofpassenger cars lead to a corresponding20% reduction in motor-fuel consumptionfor personal automotive travel? Economictheory suggests that it will not. Sinceenergy efficiency improvements reducethe marginal cost of energy services suchas travel, the consumption of thoseservices may be expected to increase. Forexample, since the cost per mile of drivingis cheaper, consumers may choose todrive further and/or more often. Thisincreased consumption of energy servicesmay be expected to offset some of thepredicted reduction in energyconsumption.

This so-called direct rebound effect wasfirst brought to the attention of energyeconomists by Daniel Khazzoom (1980)

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

1

1. Introduction

S4134 12/10/07 13:56 Page 1

and has since been the focus of muchresearch (Greening, et al., 2000). Buteven if the direct rebound effect is zero fora particular energy service (e.g. even ifconsumers choose not to drive any furtherin their fuel efficient car), there are anumber of other reasons why theeconomy-wide reduction in energyconsumption may be less than simplecalculations suggest. For example, themoney saved on motor-fuel consumptionmay be spent on other goods and servicesthat also require energy to provide. Theseso-called indirect rebound effects can take

a number of forms that are brieflyoutlined in Box 1.1. Both direct andindirect rebound effects apply equally toenergy efficiency improvements byconsumers, such as the purchase of amore fuel efficient car, and energyefficiency improvements by producers,such as the use of energy efficient motorsin machine tools.

As shown in Box 1.2, the overall oreconomy-wide rebound effect from anenergy efficiency improvement representsthe sum of these direct and indirecteffects. It is normally expressed as a

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

2

• The equipment used to improve energy efficiency (e.g. thermal insulation) will itselfrequire energy to manufacture and install and this ‘embodied’ energy consumptionwill offset some of the energy savings achieved.

• Consumers may use the cost savings from energy efficiency improvements topurchase other goods and services which themselves require energy to provide. Forexample, the cost savings from a more energy efficient central heating system maybe put towards an overseas holiday.

• Producers may use the cost savings from energy efficiency improvements toincrease output, thereby increasing consumption of capital, labour and materialsinputs which themselves require energy to provide. If the energy efficiencyimprovements are sector wide, they may lead to lower product prices, increasedconsumption of the relevant products and further increases in energy consumption.

• Cost-effective energy efficiency improvements will increase the overall productivityof the economy, thereby encouraging economic growth. The increased consumptionof goods and services may in turn drive up energy consumption.

• Large-scale reductions in energy demand may translate into lower energy priceswhich will encourage energy consumption to increase. The reduction in energy priceswill also increase real income, thereby encouraging investment and generating anextra stimulus to aggregate output and energy use.

• Both the energy efficiency improvements and the associated reductions in energyprices will reduce the price of energy intensive goods and services to a greaterextent than non-energy intensive goods and services, thereby encouragingconsumer demand to shift towards the former.

Box 1.1 Indirect rebound effects

S4134 12/10/07 13:56 Page 2

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

3

percentage of the expected energysavings from an energy efficiencyimprovement. Hence, a rebound effect of100% means that the expected energysavings are entirely offset, leading to zeronet savings.1

Rebound effects need to be defined inrelation to particular time frame (e.g.short, medium or long term) and systemboundary (e.g. household, firm, sector,national economy). The economy-widerebound effect is normally defined inrelation to a national economy, but theremay also be effects in other countriesthrough changes in trade patterns andinternational energy prices. Reboundeffects may also be expected to increasein importance over time as markets,technology and behaviour adjusts. From aclimate change perspective, it is the long-term effect on global energy consumptionthat is most relevant, but this is also theeffect that is hardest to estimate.

The economy-wide rebound effectrepresents the net effect of a number ofdifferent mechanisms that are individuallycomplex, mutually interdependent andlikely to vary in importance from one typeof energy efficiency improvement toanother. Estimating the magnitude of theeconomy-wide rebound effect in anyparticular instance is therefore verydifficult. Nevertheless, the generalimplication is that the energy ‘saved’ fromenergy efficiency improvements will beless than is conventionally assumed.

The view that energy efficiencyimprovements will increase rather thanreduce energy consumption was first putforward by the British economist, WilliamStanley Jevons, as long ago as 1865. Ithas subsequently become known as the‘Khazzoom-Brookes (K-B) postulate’, aftertwo contemporary economists (LenBrookes and Daniel Khazzoom) who havebeen closely associated with this idea.2 Inits original formulation, the K-B postulate3

states that:

‘with fixed real energy prices, energyefficiency gains will increase energyconsumption above what it would bewithout these gains’ (Saunders, 1992).

If it were correct, the K-B postulate wouldhave deeply troubling policy implications.It would imply that many of the policiesused to promote energy efficiency mayneither reduce energy consumption norcarbon emissions. The conventionalassumptions of energy analysts,policymakers, business and lay peoplealike would be turned on their head.Alternatively, even if the postulate wereincorrect, the various mechanismsdescribed above could still make energyefficiency policies less effective inreducing energy consumption than iscommonly assumed. In either case, therebound effect could have importantimplications for global efforts to addressclimate change. But despite its potentialimportance, the topic is widely neglectedand shrouded in controversy andconfusion.

1 This may be expressed as , where ENG represents the expected energy savings from a particular energy efficiencyimprovement without taking rebound effects into account; DIR represents the increase in energy consumption resulting from thedirect rebound effect; and IND represents the increase in energy consumption resulting from the indirect rebound effects.

2 The ‘Jevons-Brookes postulate’ would be a more accurate term, since Khazzoom’s work focuses entirely on the direct reboundeffect. Brookes, in contrast, is a strong advocate of the postulate at a macroeconomic level. An alternative term is ‘JevonsParadox’ (Alcott, 2005).

3 The term postulate indicates a starting assumption from which other statements are logically derived. It does not have to beself-evident or supported by empirical evidence. But since most commentators do not accept the K-B postulate, this assessmenttreats it as a hypothesis and seeks out testable implications.

S4134 12/10/07 13:56 Page 3

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

4

The economy-wide rebound effect represents the sum of the direct and indirect effects.For energy efficiency improvements by consumers, it is helpful to decompose the directrebound effect into:a) a substitution effect, whereby consumption of the (cheaper) energy service

substitutes for the consumption of other goods and services while maintaining aconstant level of ‘utility’, or consumer satisfaction; and

b) an income effect, whereby the increase in real income achieved by the energyefficiency improvement allows a higher level of utility to be achieved by increasingconsumption of all goods and services, including the energy service.

Similarly, the direct rebound effect for producers may be decomposed into:a) a substitution effect, whereby the cheaper energy service substitutes for the use of

capital, labour and materials in producing a constant level of output; and

b) an output effect, whereby the cost savings from the energy efficiency improvementallows a higher level of output to be produced - thereby increasing consumption ofall inputs, including the energy service.

It is also helpful to decompose the indirect rebound effect into:a) the embodied energy, or indirect energy consumption required to achieve the

energy efficiency improvement, such as the energy required to produce and installthermal insulation; and

b) the secondary effects that result as a consequence of the energy efficiencyimprovement, which include the mechanisms listed in Box 1.1.

A diagrammatic representation of this classification scheme is provided below (see alsothe Supplementary Note). The relative size of each effect may vary widely from onecircumstance to another and in some cases individual components of the rebound effectmay be negative. For example,if an energy service is an‘inferior good’, the incomeeffect for consumers may leadto reduced consumption ofthat service, rather thanincreased consumption. It istheoretically possible for theeconomy-wide rebound effectto be negative (‘superconservation’), although thisappears unlikely in practice.

Box 1.2 Classifying rebound effects

Secondary effects

Embodied energy

Income / outputeffect

Substitutioneffect

Actual energy savings

Indirect rebound

effect

Directreboundeffect

Economy-wide

rebound effect

‘Engineering’estimate of

energysavings

S4134 12/10/07 13:56 Page 4

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

5

Rebound effects tend to be almostuniversally ignored in official analyses ofthe potential energy savings from energyefficiency improvements. A rare exceptionis UK policy to improve the thermalinsulation of households, where it isexpected that some of the benefits will betaken as higher internal temperaturesrather than reduced energy consumption(DEFRA, 2007). But the direct reboundeffects for other energy efficiencymeasures are generally ignored, as arethe potential indirect effects for allmeasures. Much the same applies toenergy modelling studies and toindependent estimates of energyefficiency potentials by energy analysts.For example, the Stern Review of theeconomics of climate change overlooksrebound effects altogether (Stern, 2007),while the Fourth Assessment Report fromthe Intergovernmental Panel on ClimateChange simply notes that the literature isdivided on the magnitude of this effect(IPCC, 2007). Criticising this stance, theHouse of Lords Select Committee onScience and Technology commented that:

“….the Khazzoom Brookes postulate,while not proven, offers at least aplausible explanation of why in recentyears improvements in ’energy intensity‘at the macroeconomic level havestubbornly refused to be translated intoreductions in overall energy demand. TheGovernment have so far failed to engagewith this fundamental issue, appearing to

rely instead on an analogy betweenmicro- and macroeconomic effects.” (HoL,2006)4

While energy economists recognise thatdirect and indirect rebound effects mayreduce the energy savings from energyefficiency improvements, there is disputeover how important these effects are -both individually and in combination.Some argue that rebound effects are ofminor importance for most energyservices, largely because the demand forthose services appears to be inelastic5 inmost cases and because energy typicallyforms a small share of the total costs ofthose services (Lovins, et al., 1988;Lovins, 1998; Schipper and Grubb,2000).6 Others argue that they aresufficiently important to completely offsetthe energy savings from improved energyefficiency (Brookes, 2000; Herring, 2006).At first sight, it appears odd for competentand experienced analysts to hold suchwidely diverging views on what appears tobe an empirical question. This suggestsone of three things: first, different authorsmay be using different definitions of therebound effect, together with differentdefinitions of associated issues such asthe relevant system boundaries; second,the empirical evidence for rebound effectsmay be sufficiently sparse, ambiguousand inconclusive to be open to widelyvarying interpretations; or third,fundamental assumptions regarding howthe economy operates may be in dispute.

4 This is strictly incorrect. First, the K-B postulate could explain a relatively slow rate of improvement in energy intensity; andsecond, to reduce overall energy demand the rate of improvement would need to be greater than the rate of GDP growth.

5 The own-price elasticity of demand for a good or service is the ratio of the relative (or percentage) change in quantitydemanded to the relative change in price, holding other factors constant. In most cases, the own-price elasticity is negative.Demand is said to be inelastic when the own-price elasticity is less than one in absolute value, and elastic when it is greaterthan one.

6 The price of a service depends upon the total cost of providing that service, which includes capital, labour, materials and energycosts. If energy forms a small proportion of total costs, the elasticity of demand with respect to energy prices should be lessthan the own-price elasticity of demand with respect to total costs.

S4134 12/10/07 13:56 Page 5

In practice, all of these factors appear tobe relevant. The empirical evidence fordirect rebound effects is very patchy,mainly focused on a limited number ofconsumer energy services such aspersonal automotive transport and almostwholly confined to OECD economies.Quantitative estimates of indirect andeconomy-wide rebound effects are rare,leaving authors such as Brookes (2000) torely on an eclectic mix of theoreticalargument, anecdotal examples and‘suggestive’ evidence from econometricanalysis and economic history. Much ofthis evidence rests upon theoreticalassumptions and methodologicalapproaches that are both highly technicaland openly contested. There is alsoconfusion, misunderstanding anddisagreement over basic definitionalissues such as the meaning of‘improvements in energy efficiency’. Theresult is that commentators talk past oneanother, with disagreements originating inpart because different people are talkingabout different things. For example, somecommentators appear to equate ‘the’rebound effect solely with the directeffect, and thereby ignore the indirecteffects that are the primary concern ofeconomists such as Jevons. In particular,there appears to be a divergence betweenthe world views of ‘economists’ and‘engineers’:

“For the economist the world is an openfield in which different factors ofproduction and different consumptiongoods have to be constantly re-combinedin the most profitable fashion in responseto their prices and marginal productivities.An efficiency increase in one factorimmediately leads to a complete re-

arrangement due to implicit price changesand hence to a large reboundeffect..….For the engineer, instead, theworld consists of a set of giventechnologies or activities, whichdetermine demand in which relativeshares are fixed. Any increase in theproductivity of one factor affects only thatfactor and hence there is no reboundeffect.” (Birol and Keppler, 2000)

Rebound effects could therefore be ofconsiderable importance, but they arewidely ignored by policy makers and theirmagnitude is greatly disputed, evenamong economists who understand themechanisms involved. An assessment ofthe state of knowledge in this area shouldtherefore make a valuable contribution tocontemporary policy debates. At the sametime, the diversity and ambiguity of theevidence base makes such an assessmentchallenging to conduct.

1.2 How the assessmentwas conductedThis assessment is one in a series beingcarried out by the Technology and PolicyAssessment (TPA) function of the UKEnergy Research Centre (UKERC). Theselection of the rebound effect as a topicfollowed wide ranging consultation withstakeholders, together with advice fromthe TPA Advisory Group. The Group notedthe persistence of controversy about therebound effect, the existence of widelydiverging views on the topic and themismatch between the potentialimportance of the topic and theapparently limited research devoted to it.It also emphasised the importance of

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

6

S4134 12/10/07 13:56 Page 6

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

7

addressing rebound effects ‘in the round’,including the indirect and economy-wideeffects. This led the UKERC to undertakethis study.

The objective of this assessment is not toundertake new research on reboundeffects. Instead, it is to provide athorough review of the current state ofknowledge on this issue and to explain thefindings in an accessible manner. Thegeneral approach was informed by thesystematic review techniques prominentin medicine and other fields (see Box 1.3).Following this model, the assessmentbegan with a Scoping Note that clarifiedthe definition and nature of the reboundeffect, identified the size and nature of theevidence base and set out the options forsynthesising this evidence (Sorrell and

Dimitropoulos, 2005b). The diversity ofthe evidence base and the extent to whichit is embedded in complex and contestedtheoretical issues was found to make theconventional systematic reviewtechniques difficult to apply – a featurethat is common to many policy-relevantquestions in the energy field (Sorrell,2007).

An Advisory Group for the project wasestablished and the Scoping Notecirculated to key stakeholders. This led toa series of recommendations on theappropriate scope and focus of theassessment, including the relative weightto be given to different sources ofevidence. The agreed approach was thenset out in an Assessment Protocol (Sorrelland Dimitropoulos, 2005a) which clarifiedthat the assessment should update and

The TPA approach to assessment seeks to learn from a range of techniques referred toas Evidence Based Policy and Practice (EBPP), including meta-analyses of quantitativedata and systematic reviews of quantitative and qualitative evidence. These aspire toprovide more robust evidence for policymakers and practitioners, avoid duplication ofresearch, encourage higher research standards and identify research gaps. However,energy policy presents a number of challenges for the application of systematic reviewtechniques and the approach has been criticised for excessive methodological rigidity insome policy areas (Sorrell, 2007). The UKERC has therefore set up a process that isinspired by these approaches, but is not bound to any narrowly defined method ortechnique.

The process carried out for each assessment includes the following components:• Publication of Scoping Note and Assessment Protocol.• Establishment of a project team with a diversity of expertise.• Convening an Expert Group with a diversity of opinion and perspective.• Stakeholder consultation.• Systematic searches of the the evidence base.• Categorisation and assessment of evidence.• Synthesis, review and drafting.• Expert feedback on initial drafts.• Peer review of final draft.

Box 1.3 Overview of the TPA approach

S4134 12/10/07 13:56 Page 7

extend a previous literature review byGreening et al. (2000), place greateremphasis on indirect and economy-wideeffects and focus in particular on clarifyingthe underlying conceptual frameworksand identifying the reasons for the widelydiverging views. Since disputes over therebound effect involve some highlytechnical concepts from economic theoryand econometric practice, much emphasishas been placed on clarifying theseconcepts and making them accessible to anon-technical audience.

The assessment was organised aroundfive broad categories of evidence, asfollows:

• Evaluation studies: micro-levelevaluations of the impact of specificenergy efficiency improvements onthe demand for energy or energyservices;

• Econometric studies: use ofsecondary data sources to estimatethe elasticity of the demand forenergy or energy services at differentlevels of aggregation;

• Elasticity of substitution studies:estimates of the elasticity ofsubstitution between energy andcapital at different levels ofaggregation;

• Computable general equilibriummodelling studies: estimates ofeconomy-wide rebound effects fromcomputable general equilibrium (CGE)models of the macroeconomy; and

• Energy, productivity and economicgrowth studies: a range of evidenceconsidered in some way relevant tothe K-B postulate, including studiesfrom economic history, neoclassical

production theory, neoclassical growththeory, ecological economics,decomposition analysis, and input-output analysis.

The first two categories of evidence relateto direct rebound effects while theremainder relate to economy-widerebound effects. The last category isespecially diverse and has received thegreatest amount of attention. While theproject began with systematic literaturesearches using keyword such as ‘reboundeffect’, the final assessment incorporatesa much broader range of evidence,reviewing more than 500 studies fromaround the world. The full results of theassessment are reported in five in-depthTechnical Reports, namely:

• Technical Report 1: Evidence fromevaluation studies

• Technical Report 2: Evidence fromeconometric studies

• Technical Report 3: Evidence fromelasticity of substitution studies

• Technical Report 4: Evidence fromCGE modeling studies

• Technical Report 5: Evidence fromenergy, productivity and economicgrowth studies

In addition, a shorter Supplementary Noteprovides a graphical analysis of reboundeffects. All these reports are available todownload from the UKERC website.

The present report summarises the mainconclusions of the assessment andidentifies policy implications and prioritiesfor further research.

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

8

S4134 12/10/07 13:56 Page 8

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

9

1.3 Report structureThe report is structured as follows:

Section 2 describes the differentdefinitions of energy efficiency, thechoices available for the independent anddependent variables for the reboundeffect and the implications of thosechoices for the estimated magnitude ofthe effect. It is argued that disputes overthe magnitude of rebound effects result inpart from confusion over these basicdefinitions.

Section 3 describes the nature andoperation of direct rebound effects andsummarises the evidence regarding themagnitude of these effects for a numberof consumer energy services. It concludesthat direct rebound effects appear to below to moderate (i.e. <30%) for mostconsumer energy services in developedcountries. Direct rebound effects may belarger in developing countries and forenergy efficiency improvements byproducers, but the empirical evidence forboth is weak.

Section 4 describes the nature andoperation of indirect and economy-widerebound effects and summarises theresults of a number of studies that providequantitative estimates of these effects.This evidence base is both small andsubject to a number of importantmethodological weaknesses, making it

difficult to draw any general conclusions.However, the available studies suggestthat economy-wide rebound effects mayfrequently exceed 50%.

Section 5 summarises the evidence forthe K-B postulate, focusing in particularon the work of W.S. Jevons, Len Brookesand Harry Saunders. Instead of providingquantitative estimates of rebound effects,the evidence discussed in this sectionprovides indirect support for the K-Bpostulate and is taken from such diversefields as neoclassical growth theory,economic history and econometricanalysis of productivity trends. A coreargument of this section is that the casefor the K-B postulate hinges on the claimthat energy plays a more important role ineconomic growth than is conventionallyassumed. Much of the discussiontherefore focuses on this broader issueand compares conventional views with thealternative perspective of ecologicaleconomics. It is argued that satisfactoryresolution of the debate over the K-Bpostulate may hinge in part on asatisfactory resolution of this muchbroader question.

Finally, Section 6 summarises the keyconclusions from the assessment,identifies the main research needs andhighlights some important policyimplications.

S4134 12/10/07 13:56 Page 9

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

10

S4134 12/10/07 13:56 Page 10

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

11

2. What is energy efficiency?

This section introduces the differentdefinitions of energy efficiency and thedifferent ways of measuring energy inputsand useful outputs. It argues that thesedefinitional issues have important butgenerally unacknowledged implicationsfor the estimated magnitude of reboundeffects.

2.1 Energy efficiency andenergy savingsEnergy efficiency improvements aregenerally assumed to reduce energyconsumption below where it would havebeen without those improvements. Therebound effect may reduce the size ofthese energy savings (Box 1.2). However,estimating the size of any ‘energy savings’is far from straightforward, since:

• real-world economies do not permitcontrolled experiments, so therelationship between a change inenergy efficiency and a subsequentchange in energy consumption islikely to be mediated by a host ofconfounding variables;

• we can’t observe what energyconsumption ‘would have been’without the energy efficiencyimprovement (the so-called‘counterfactual’ scenario), so theestimated ‘savings’ from energyefficiency improvement will always beuncertain; and

• energy efficiency is not controlledexternally by an experimenter andmay be influenced by a variety oftechnical, economic and policyvariables. In particular, the directionof causality may run in reverse - with

changes in energy consumption(whatever their cause) leading tochanges in energy efficiency.

Energy efficiency may be defined as theratio of ‘useful’ outputs to energy inputsfor a system. The system in question maybe an individual energy conversion device(e.g. a boiler), a building, an industrialprocess, a firm, a sector or an entireeconomy. In all cases, the measure ofenergy efficiency will depend upon how‘useful’ is defined and how inputs andoutputs are measured. When outputs aremeasured in thermodynamic or physicalterms, the term energy efficiency tends tobe used, but when outputs are measuredin economic terms it is more common touse the term ‘energy productivity’. Theinverse of both measures is termed‘energy intensity’.

Improvements in energy efficiency mayreduce the energy used by that system.For example, the installation of acondensing boiler may reduce the amountof gas used to heat a house. But a fullaccounting of rebound effects requiresattention to be paid to the consequencesfor energy consumption within broadersystems, such as the economy as a whole.For example, the money saved on heatingcosts may be spent on other goods andservices which also require energy toprovide.

The definition of inputs and outputs, theappropriate system boundaries formeasures of energy efficiency and energyconsumption and the timeframe underconsideration can vary widely from onestudy to another. The conclusions drawnregarding the magnitude and importanceof the rebound effect are likely to dependupon the particular choices that are made.

S4134 12/10/07 13:56 Page 11

2.2 Multiple definitionsThe most basic definition of energyefficiency derives from the first-law ofthermodynamics and measures the ratioof ‘useful’ energy outputs to the heatcontent, or calorific value of fuel inputs. Aconventional lightbulb, for example, has afirst-law efficiency of only 6%, since 6%of the heat content of energy inputs areconverted to light energy and theremainder is lost as ‘waste’ heat (Berndt,1978; Patterson, 1996). But the first-lawefficiency of a process depends upon how‘useful’ is defined. When waste heat andother losses are taken into account, thefirst-law efficiency becomes 100%, sinceenergy is not ‘used up’ but is merelytransformed from ‘available’ to lessavailable forms.

Although widely used, first-law efficiencymeasures are misleading because they donot take into account the availability ofenergy inputs and outputs, or their abilityto perform useful work7 (Berndt, 1978).For example, energy in the form of high-pressure steam can perform more usefulwork than the same amount of energy inthe form of low temperature heat. Thethermodynamic concept of exergyprovides a general measure of the abilityto perform useful work and may beapplied to both the inputs and outputs ofconversion processes (Wall, 2004). Theexergy content of an energy carrier maytherefore be different from its heatcontent, although both are measured inkWh (kilowatt hours). For example, a heatunit of electricity will be ranked higher on

exergy basis than a heat unit of oil ornatural gas. Also, unlike energy, exergy is‘consumed’ in conversion processes.

The notion of exergy leads to a seconddefinition of energy efficiency, based uponthe second law of thermodynamics. This‘second-law’ measure is frequentlysmaller than the first-law efficiency,suggesting a greater potential forimprovement. For example, the first-lawefficiency of electric resistance spaceheating may exceed 99%, but this falls toaround 5% when a second-law definitionis used (Rosen, 2004). The differencearises because resistance heatingconverts high exergy electricity to lowexergy space heat. Second-law measuresare preferable since they focus attentionwhat needs to be conserved - namelyexergy rather than energy per se (Berndt,1978).

For most purposes, it is simpler tomeasure useful energy outputs in terms ofphysical indicators for the relevant energyservice, rather than heat content orexergy. For example, a suitable outputmeasure for personal transportation byprivate car could be vehicle kilometres. Aphysical measure of energy efficiencywould then be vehicle kilometres per litreof motor fuel.

Physical measures may be applied at thelevel of individual energy conversiondevices, but are more commonly appliedat higher levels of aggregation, such as ahousehold, an industrial process, anindividual firm or an individual sector. Ineach case, changes in physical measures

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

12

7 ‘Work’ may be broadly defined as an increase in the kinetic, potential, physical or chemical energy of a subsystem that islocated within a larger system in which energy - according to the first-law - is always conserved (Ayres, et al., 2003).

S4134 12/10/07 13:56 Page 12

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

13

may result from factors other thanimprovements in the thermodynamicefficiency of conversion devices. Forexample, changes in vehicle load factorswill change a physical measure of energyefficiency that is based upon passengerkilometres per litre of motor-fuel.Appropriate physical indicators are likelyto vary from one sector to another andone type of energy service to another,making an aggregate economy-widephysical measure of energy efficiencyinappropriate.

By replacing the numerator with anindicator of the economic value of output,the energy efficiency of different sectorscan be compared (Patterson, 1996). Forexample, the energy efficiency of both thebrewing and dairy sectors can bemeasured in terms of value added perGigawatt hours (GWh) of energy input.The move from physical to economicindicators increases the number of factorsthat could influence the indicator as doesthe use of such indicators at higher levelsof aggregation. The indicator that isfurthest from a thermodynamic measureof energy efficiency is therefore the ratioof GDP to total primary energyconsumption within a national economy.

Hence, when using the terms energyefficiency and energy productivity, it isessential to clarify how the inputs andoutputs are being measured. In whatfollows, the term useful work will be usedin a generic sense to refer to the usefuloutputs of energy conversion systems.

Whichever measure is used, maximisingenergy efficiency is an inappropriate goalsince it does not take into account thecosts associated with other inputs such as

capital and labour. Economists aretherefore more concerned with improvingtotal factor productivity (TFP), which is ameasure of the efficiency with which allinputs are used by a firm, sector ornational economy.

The energy efficiency improvements thatare of most relevance to the reboundeffect are those that are consistent withthe best use of all economic resources.These are conventionally divided into twocategories: those that are associated withimprovements in total factor productivity(‘technical change’), and those that arenot (‘substitution’). The former areconventionally assumed to occurindependently of changes in relativeprices, while the latter are assumed tooccur in response to such changes (Box2.1). This distinction is potentiallymisleading, however, since it assumesthat the producer is making fully efficientuse of all relevant inputs. Many energyefficiency improvements may be betterdescribed as ‘overcoming inefficiency’,since they represent the use of existingtechnologies to move closer to an‘optimal’ combination of inputs (Box 2.1).This type of improvement also improvestotal factor productivity and mayfrequently be stimulated through non-price energy efficiency policies.

Rebound effects may be defined inrelation to each type of energy efficiencyimprovement, although most studiesclassify such improvements as derivingfrom either substitution or technicalchange. The consequences of technicalchange are of particular interest, sincethis contributes to the growth in economicoutput. Also, many (if not most) energyefficiency improvements are likely to be

S4134 12/10/07 13:56 Page 13

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

14

Increases in energy prices may lead to improvements in energy efficiency (or energyproductivity) of a system through the substitution of capital or labour inputs for energy.For example, thermal insulation may reduce the consumption of gas for space heating.But if the prices of other inputs are unchanged the total cost of producing a given levelof output will have increased. In contrast, improvements in both energy productivity andtotal factor productivity may result from technical change. This is conventionallyassumed to occur independently of any change in relative prices and is considereddesirable since it occurs without any reduction in economic output.

In neoclassical economics, a production function represents the maximum possible flowof output (Y) obtainable from the flow of energy (E) and other inputs (X), given thecurrent state of technology (see diagram). An ‘isoquant’ of a production functionrepresents the different possible combination of inputs that may be used to produce agiven level of output. The ‘optimal’ combination depends upon the relative prices ofthose inputs. Substitution may be represented a movement along an isoquant inresponse to a change in relative prices. This may require investment in technologies that

can combine inputs indifferent ways (e.g. energy-efficient motors), but theseare assumed to be chosenfrom a set of existingtechnologies. In contrast,technical change refers to thedevelopment of newtechnologies and methods oforganisation that shift theisoquant to the left, allowingthe same level of output to beproduced from a lower levelof inputs.

In practice, the conventional distinction between substitution and technical change canbe misleading. First, the notion of substitution implies a ‘frictionless’ move from oneexisting technique to another, but in practice this requires investment and will take time.Second, technical change is not autonomous but is influenced by changes in relativeprices and much contemporary research is concerned with improving understanding ofthis process (Grubb, et al., 2002). Third, the production function represents the mostefficient combination of factor inputs, but in practice firms may use relatively inefficientcombinations. Much investment may be better represented as a move from less efficientto more efficient combinations, while still remaining within the production function‘frontier’ - indicated by the ‘overcoming inefficiency’ arrow in the diagram.

Box 2.1 Types of energy efficiency improvement

Energy - E

Other inputs - X

Overcoming inefficiency

SubstitutionTechnicalchange

Y

Y

S4134 12/10/07 13:56 Page 14

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

15

the by-product of attempts to improvetotal factor productivity, rather than theresult of targeted efforts to improveenergy efficiency.

Technical change is said to be ‘neutral’ if itreduces the use of all inputs by an equalamount and ‘biased’, if it reduces the useof some inputs more than others. ‘Energy-saving’ technical change reduces theshare of energy in the value of output byproportionately more than the share ofother inputs, while ‘energy-using’technical change does the reverse. Thisbias in technical change is closely relatedto (but not the same as) the rate ofgrowth of energy efficiency over time,holding relative prices constant, which isan important parameter in many energy-economic models (Löschel, 2002;Sanstad, et al., 2006).

2.3 UnacknowledgedimplicationsThe importance of these definitionalissues for estimates of the rebound effectis not often recognised. Manycommentators assume that the relevantindependent variable for the reboundeffect is improvements in thethermodynamic efficiency of individualconversion devices or industrialprocesses. But such improvements willonly translate into comparableimprovements in different measures ofenergy efficiency, or measures of energyefficiency applicable to wider systemboundaries, if several of the mechanismsresponsible for the rebound effect fail tocome into play. For example,improvements in the number of litres

used per vehicle kilometre will onlytranslate into improvements in thenumber of litres used per passengerkilometre if there are no associatedchanges in average vehicle load factors.

Rebound effects may be expected toincrease over time and with the wideningof the system boundary for the dependentvariable (energy consumption). Forexample, the energy savings formanufacturing as a whole may beexpected to be less than the energysavings for an individual firm that investsin an energy efficient technology. For theK-B postulate the relevant systemboundary is normally taken as thenational economy. But energy efficiencyimprovements may also affect tradepatterns and international energy prices,thereby changing energy consumption inother countries. For the purpose ofassessing the contribution of energyefficiency to reducing carbon emissions,the relevant system boundary is the wholeworld. To capture the full range ofrebound effects, the system boundary forthe independent variable (energyefficiency) should be relatively narrow,while the system boundary for thedependent variable (energy consumption)should be as wide as possible. Forexample, the independent variable couldbe the energy efficiency of an electricmotor, while the dependent variable couldbe economy-wide energy consumption.However, measuring or estimating theeconomy-wide effects of such micro-levelchanges effects is, at best, challenging.For this reason, the independent variablefor many theoretical and empirical studiesof rebound effects is a physical oreconomic measure of energy efficiency

S4134 12/10/07 13:56 Page 15

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

16

that is applicable to relatively wide systemboundaries – such as the energy efficiencyof an industrial sector. But such studiesmay overlook the ‘lower-level’ reboundeffects resulting from improvements inphysical or thermodynamic measures ofenergy efficiency appropriate to narrowersystem boundaries. For example,improvements in the energy efficiency ofelectric motors in the engineering sectormay lead to rebound effects within thatsector, with the result that the energyintensity of that sector is reduced by lessthan it would be in the absence of sucheffects. But if the energy intensity of thesector is taken as the independentvariable, these lower-level reboundeffects will be overlooked. Also,improvements in more aggregatemeasures of energy efficiency are unlikelyto be caused solely (or even mainly) bythe diffusion of more thermodynamicallyefficient conversion devices.

Aggregate measures of energy efficiencywill depend upon how different types ofenergy input are combined. The mostcommon approach is to aggregatedifferent energy types according to theirheat content, but this neglects the‘quality’ of each energy type, or the abilityto perform useful work. The latter, in turn,is only one of several factors thatdetermine the economic productivity ofdifferent energy types, with othersincluding cleanliness, ameanability tostorage, safety, flexibility of use and soon, together with the use to which theenergy is put (Cleveland, et al., 2000). Inthe absence of significant marketdistortions, the relative price per kilowatt

hour of different energy carriers canprovide a broad indication of their relativemarginal productivities (Kaufmann,1994).8

Using data on relative prices, economistshave devised methods for aggregatingenergy types of different marginalproductivities (Berndt, 1978). In general,when the ‘quality’ of energy inputs areaccounted for, aggregate measures ofenergy efficiency are found to beimproving more slowly than is commonlysupposed (Hong, 1983; Zarnikau, 1999;Cleveland, et al., 2000). For example, ona thermal input basis, per-capita energyconsumption in the US residential sectordecreased by 20% over the period 1970to 1991, but when adjustments are madefor changes in energy quality (notably theincreasing use of electricity), per capitaenergy consumption is found to haveincreased by 7% (Zarnikau, et al., 1996).This difference demonstrates thattechnical progress in energy use is notconfined to improvements inthermodynamic efficiency, but alsoincludes the substitution of low qualityfuels by high quality fuels (notablyelectricity), thereby increasing theamount of utility or economic outputobtained from the same heat content ofinput (Kaufmann, 1992).

Failure to allow for this can lead tomisleading conclusions. For example, it iscommonly assumed that a combination ofstructural change and ‘energy efficiency’improvements have allowed OECDcountries to decouple GDP growth fromthe growth in primary energy

8 The marginal product of an energy input into a production process is the marginal increase in the value of output produced bythe use of one additional heat unit of energy input.

S4134 12/10/07 13:56 Page 16

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

17

consumption (Geller, et al., 2006).9

However, if energy inputs are weighted bytheir relative marginal productivities, thegrowth in energy consumption is shown tobe closely coupled to the growth in GDP(Hong, 1983; Stern, 1993; Cleveland, etal., 2000; Stern and Cleveland, 2004).Hence, not only may such ‘decoupling’ bepartly illusory, but the contribution ofimproved thermodynamic efficiency toany decoupling may easily be overstated.

Both fuel switching and improvements inthe thermodynamic efficiency ofconversion devices may be expected tochange aggregate measures of energyefficiency, but the rebound effectsassociated with each may be different.Also, improvements in any measure ofenergy efficiency rarely occur in isolationbut are typically associated with broaderimprovements in the productivity of otherinputs, with new technologies frequentlyproviding both. Pye and McKane (1998),for example, show how the installation ofenergy efficient motors can reduce wearand tear, extend the lifetime of systemcomponents and achieve savings incapital and labour costs that exceed thereduction in energy costs. Hence, if thefull impact on energy consumption of anew energy-efficient technology is takenas the appropriate dependent variable,then the estimated (direct, indirect oreconomy-wide) rebound effect may belarger. Conversely, if only a portion of thisimpact is attributed specifically to theenergy efficiency improvement, then theestimated rebound effect may be smaller.However, it is both difficult and misleading

to isolate the impact on energy demand ofan improvement in energy efficiencyalone. What matters for policy is how anew, energy-efficient technology affectsoverall energy demand. This frequentlyoverlooked point is a major theme of thisassessment.

2.4 Summary• Energy efficiency may be measured in

a variety of ways for a variety ofsystem boundaries and any one ofthese may form an appropriateindependent variable for an estimateof the rebound effect. While theappropriate choice depends upon theobjectives of the study, difficultiesmay arise if commentators interpretthe term ‘energy efficiency’ indifferent ways.

• The dependent variable for therebound effect is a change in energyconsumption, which may also bemeasured for a variety of systemboundaries. However, wider systemboundaries allow a greater range ofrebound effects to be captured.

• Many theoretical and empirical studiesof rebound effects employ physical oreconomic measures of energyefficiency applicable to relatively widesystem boundaries. But these mayoverlook the rebound effects resultingfrom improvements in physical orthermodynamic measures of energyefficiency appropriate to narrowersystem boundaries.

9 If energy is measured in terms of heat content, major OECD countries used one third less primary energy to generate a unitof GDP than in the early 1970s (Geller, et al., 2006).

S4134 12/10/07 13:56 Page 17

• Many theoretical and empirical studiesaggregate different types of energycarrier on the basis of their thermalcontent, thereby neglectingdifferences in energy quality. This maylead such studies to overlook thecontribution of changes in fuel mix tochanges in aggregate measures ofenergy efficiency. If energy inputs areweighted by their relative marginalproductivities, the historical growth inenergy consumption appears to beclosely coupled to the growth in GDP.

• Economists are primarily interested inenergy efficiency improvements thatare consistent with the best use of alleconomic resources. These arecommonly categorised as either price-induced substitution or technicalchange, but this neglects the

importance of organisational andmarket failures in contributing toinefficiencies as well as the role ofrelative prices in stimulating technicalchange.

• Energy efficiency improvements byproducers may often be associatedwith improvements in the productivityof capital, labour and materialsinputs. Similarly, energy efficiencyimprovements by consumers mayoften save costs on more than energyalone. This needs to be allowed forwhen estimating rebound effects, asdoes any associated changes in fuelmix. Failure to do so may lead torebound effects being underestimatedand to the potential for decouplingbetween energy consumption andeconomic output to be overestimated.

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

18

S4134 12/10/07 13:56 Page 18

This section summarises the empiricalevidence for direct rebound effects,focusing in particular on energy servicesin the household sector, since this iswhere the bulk of the research has beenundertaken. A full examination of thisevidence is contained in Technical Report1 and Technical Report 2.

Section 3.1 describes the operation of thedirect rebound effect, highlighting somekey issues concerning the measurementof this effect and the conditions underwhich it may be expected to be larger orsmaller. The Supplementary Note containsa simple graphical analysis of reboundeffects that complements the discussionin this section.

Section 3.2 describes one approach toestimating direct rebound effects that usequasi-experimental methodologiesadopted from the field of policyevaluation. It then summarises the resultsof a limited number of studies that usethis approach to estimate direct reboundeffects from energy efficiencyimprovements in space heating.

Section 3.3 describes an alternative andmore commonly used approach toestimating direct rebound effects thatuses econometric analysis of secondarydata sources to estimate the elasticity ofdemand for either useful work or energyconsumption. Sections 3.4 and 3.5summarise the results of a number ofstudies that use this approach to estimatedirect rebound effects for personalautomotive transport, household heatingand a limited number of other consumerenergy services. Section 3.6 discusses anumber of potential sources of bias withthis approach that may lead the directrebound effect to be overestimated.Section 3.7 concludes.

3.1 Understanding thedirect rebound effectDirect rebound effects relate to individualenergy services, such as heating, lightingand refrigeration and are confined to theenergy required to provide that service.Improved energy efficiency will decreasethe marginal cost of supplying that serviceand should therefore lead to an increasein consumption of the service. Forexample, consumers may choose to drivefurther following the purchase of energyefficient car because the price perkilometre has fallen. The resultingincrease in energy service consumptionwill tend to offset the expected reductionin energy consumption provided by theenergy efficiency improvement.

As shown in Box 1.2, the direct reboundeffect for consumers may be decomposedinto a substitution effect, wherebyconsumption of the (cheaper) energyservice substitutes for the consumption ofother goods and services whilemaintaining a constant level of ‘utility’;and an income effect, whereby theincrease in real income achieved by theenergy efficiency improvement allowsincreased consumption of the energyservice. In a similar manner, the directrebound effect for producers may bedecomposed into a substitution effect andan output effect (see SupplementaryNote). Also, for most energy services, thedirect rebound effect may be expected toincrease over time as markets,technologies and behaviours adjust. Forexample, energy efficiency improvementsmay lower production costs for a firm, butit may take time for the firm to increaseoutput and market share.

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

19

3. The evidence for direct rebound effects

S4134 12/10/07 13:56 Page 19

Energy services are provided through acombination of capital equipment, labour,materials and energy. An essential featureof an energy service is the useful workobtained which, as shown in Section 2,may be measured by a variety of ways.For example, the useful work frompassenger cars may be measured invehicle kilometres, passenger kilometresor (rather unconventionally) in tonnekilometres. Energy services may alsohave broader attributes that may becombined with useful work in a variety ofways. For example, all cars deliverpassenger kilometres, but they may varywidely in terms of features such as speed,comfort, acceleration and prestige.Consumers and producers may thereforemake trade-offs between useful work andother attributes of an energy service;between energy, capital and other marketgoods in the production of an energyservice; and between different types ofenergy service.

By reducing the marginal cost of usefulwork, energy efficiency improvementsmay, over time, lead to an increase in thenumber of energy conversion devices,their average size, their averageutilisation and/or their average loadfactor. For example, people may buy morecars, buy larger cars, drive them furtherand/or share them less. Similarly, peoplemay buy more washing machines, buylarger machines, use them morefrequently and/or reduce the size of theaverage load. The relative importance ofthese variables may be expected to varywidely between different energy servicesand over time. For example, technologicalimprovements in the energy efficiency ofnew refrigerators are unlikely to increase

the average utilisation of the refrigeratorstock (measured in hours/year) but couldlead to a long-term increase in both thenumber of refrigerators and their averagesize (since the cost per cubic metre ofrefrigeration has fallen). Over the verylong-term, the lower cost of energyservices may contribute to fundamentalchanges in technologies, infrastructuresand lifestyles - such as a shift towardscar-based commuting and increasingdistances between residential, workplaceand retail locations. But as the timehorizon extends, the effect of suchchanges on the demand for the energyservice (as well as the demand for othergoods and services) becomes increasinglydifficult to separate from the effect ofincome growth and other factors.

The estimated size of the direct reboundeffect depends upon how useful work andhence energy efficiency is defined. Forexample, the majority of estimates of thedirect rebound effect for personalautomotive transport measure usefulwork in terms of vehicle kilometrestravelled, which is sometimesdecomposed into the product of thenumber of vehicles and the mean distancetravelled per vehicle per year (Greene, etal., 1999b; Small and Van Dender, 2005).Energy efficiency is then defined asvehicle kilometres per litre of fuel andrebound effects are measured asincreases in distance driven. But thisoverlooks any changes in mean vehiclesize and weight as a result of energyefficiency improvements (e.g. moreSUVs), as well as any decrease in averagevehicle load factor (e.g. less car sharing).If energy efficiency was measured insteadas tonne kilometres per litre of fuel,

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

20

S4134 12/10/07 13:56 Page 20

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

21

rebound effects would show up as anincrease in tonne kilometres driven, whichmay be decomposed into the product ofthe number of vehicles, the mean vehicleweight and the mean distance travelledper vehicle per year. To the extent thatvehicle weight provides a proxy for factorssuch as comfort, safety and carryingcapacity, this approach effectivelyincorporates some features normallyclassified as attributes of the energyservice into the measure of useful work. Italso moves closer to a thermodynamicmeasure of energy efficiency, by focusingupon the movement of mass rather thanthe movement of people.

The magnitude of direct rebound effectsmay be expected to be proportional to theshare of energy in the total cost of energyservices10, as well as the extent to whichthose costs are ‘visible’ to the consumer.11

But as the consumption of a particularenergy service increases, saturationeffects (technically, declining marginalutility) should reduce the size of any directrebound effect. For example, directrebound effects from improvements in theenergy efficiency of household heatingsystems should decline rapidly oncewhole-house indoor temperaturesapproach the maximum level for thermalcomfort. One important implication is thatdirect rebound effects will be higheramong low income groups, since theseare further from satiation in theirconsumption of many energy services(Boardman and Milne, 2000).

Increases in demand for an energy servicemay derive from existing consumers ofthe service, or from consumers who werepreviously unable or unwilling to purchasethat service. For example, improvementsin the energy efficiency of home air-conditioners may encourage consumers topurchase portable air-conditioners for thefirst time. The abundance of such‘marginal consumers’ (Wirl, 1997) indeveloping countries points to thepossibility of large rebounds in thesecontexts, offset to only a limited extent bysaturation effects among existingconsumers (Roy, 2000).

While energy efficiency improvementsreduce the energy cost of energy services,the size of the direct rebound effect willdepend upon how other costs areaffected. For example, direct reboundeffects may be smaller if energy efficientequipment is more expensive than lessefficient alternatives, because suchimprovements should not encourage anincrease in the number and capacity ofconversion devices. However, oncepurchased, such devices may be expectedto have a higher utilisation. In practice,many types of equipment appear to haveboth improved in energy efficiency overtime and fallen in total cost relative toincome.

Even if energy efficiency improvements

are not associated with changes in capital

or other costs, certain types of direct

rebound effect may be constrained by the

10 For example, if energy accounts for 50% of the total cost of an energy service, doubling energy efficiency will reduce thetotal costs of the energy service by 25%. But if energy only accounts for 10% of total costs, doubling energy efficiency will reducetotal cost by only 5%. In practice, improvements in energy efficiency may themselves be costly.

11 For example, Kempton and Montgomery (1982) have compared the information value of the average household energy billto that of receiving a single monthly bill from the supermarket for ‘food’. Recent developments in smart metering and electronicdisplay technologies offer the opportunity to improve the information available to consumers, and this could potentially makethe demand for consumer energy services more price elastic.

S4134 12/10/07 13:56 Page 21

real or opportunity costs associated with

increasing demand. Two examples are the

opportunity cost of space (e.g. increasing

refrigerator size may not be the best use

of available space) and the opportunity

cost of time (e.g. driving longer distances

may not be the best use of available

time). However, space constraints may

become less important over time if

technological improvements reduce the

average size of conversion devices per

unit of output or if rising incomes lead to

an increase in average living space (e.g.

compare refrigerator sizes in the US and

the UK) (Wilson and Boehland, 2005). In

contrast, the opportunity cost of time

should increase with rising incomes. As

with energy price elasticities, therefore,

the direct rebound effect for a particular

energy service may be expected to vary

over time and to be influenced by income

and other variables.

3.2 Estimating directrebound effects fromevaluation studiesOne approach to estimating the direct

rebound effect is to measure the change

in demand for useful work following an

energy efficiency improvement: for

example, measuring the change in

internal temperatures following the

installation of a fuel-efficient boiler. The

demand for useful work before the energy

efficiency improvement could be taken as

an estimate for what demand ‘would have

been’ in the absence of the improvement.

However, various other factors may also

have changed the demand for useful work

which needs to be controlled for.

Since it can be very difficult to measure

useful work for many energy services

(Box 3.1), an alternative approach is to

measure the change in energy

consumption for that service following an

energy efficiency improvement. But to

estimate direct rebound effects, this

needs to be compared with a

counterfactual scenario for energy

consumption that has at least two sources

of error, namely: a) the energy

consumption that would have occurred

without the energy efficiency

improvement; and b) the energy

consumption that would have occurred

following the energy efficiency

improvement had there been no

behavioural change. The first of these

gives an estimate of the energy savings

from the energy efficiency improvement,

while the second isolates the rebound

effect. Estimates for the latter can be

derived from engineering models, but

these frequently require data on the

circumstances of individual installations

and are prone to error.

Both these approaches are analogous to

the policy evaluation strategies employed

in areas such as health and labour

economics, but these appear to be

relatively rare in the energy field owing in

part to measurement difficulties (Frondel

and Schmidt, 2005). There are relatively

few published studies and nearly all of

these studies focus on consumer energy

services. Nadel (1993) reports the results

of a number of evaluation studies by US

utilities, which suggest direct rebound

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

22

S4134 12/10/07 13:56 Page 22

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

23

effects of 10% or less for lighting and

approximately zero for water heating,

with inconclusive results for refrigeration.

We have not been able to access these

studies, which appear to be small-scale,

short-term and methodologically weak.

Instead, we summarise the results of 15

studies of household heating.

These studies use a variety of approaches

to study a range of energy efficiency

improvements for different types of

household and different income groups,

mostly within the US or UK. The

methodological quality of most of these

studies is relatively poor, with the majority

using simple before and after

comparisons, without the use of a control

group or explicitly controlling for

confounding variables. This is the weakest

evaluation strategy and prone to bias

(Meyer, 1995; Frondel and Schmidt,

2005). Also, several studies are subject to

selection bias, since households choose to

participate rather being randomly

assigned (Hartman, 1988). While

techniques are available to control for

this, they are rarely used (Berry, 1983;

Train, 1994). Other methodological

weaknesses include: small sample sizes

contributing to low statistical power;12

high variance in results and a frequent

failure to present the error associated

with estimates; large variation in the

relevant independent variable both within

and between studies (e.g. participating

households receiving different types of

energy efficiency measure, or

combinations of measures); monitoring

periods that are too short (e.g. one

month) to capture either long-term

behavioural changes or seasonal

variations in behavioural response; and so

on. These weaknesses reduce the degree

of confidence in the results and create

difficulties in comparing the results of

different studies.

For household heating, it is helpful to

distinguish between:

• shortfall, representing the difference

between actual savings in energy

consumption and those expected on

the basis of engineering estimates;

• temperature take-back, representing

the change in internal temperature

following the energy efficiency

improvement; and

• behavourial change, representing the

proportion of change in internal

temperature that derives from

adjustments of heating controls and

other variables by the user (e.g.

opening windows).

Typically, only a portion of temperature

take-back is due to behavioural change,

with the remainder being due to physical

and other factors (Sanders and Phillipson,

2006).13 Similarly, only a portion of

shortfall may be due to temperature take-

back, with the remainder being due to

12 The power of a statistical test is the probability that the test will reject a false null hypothesis.

13 For example, daily average household temperatures will generally increase following improvements in thermal insulation,even if the heating controls remain unchanged. This is because insulation contributes to a more even distribution of warmtharound the house, reduces the rate at which a house cools down when the heating is off and delays the time at which it needsto be switched back on (Milne and Boardman, 2000).

S4134 12/10/07 13:56 Page 23

poor engineering estimates of potential

savings, inadequate performance of

equipment, deficiencies in installation and

so on.14 Hence, behavioural change is

one, but not the only (or necessarily the

most important) explanation of

temperature take-back and the latter is

one, but not the only explanation of

shortfall. Several studies misleadingly

equate shortfall with behavioural change

and fail to specify the relevant

uncertainties.

Direct rebound effects are normallyinterpreted as a behavioural response tolower energy service prices and hencemay be best approximated by thebehavioural change identified above. Butit may be misleading to interpret thischange solely as a rational response tolower heating costs, partly becauseenergy efficiency improvementsfrequently change other variables (e.g.airflow) that also encourage behaviouralresponses. Also, measures of temperaturetake-back may be difficult to translate intoestimates of shortfall because of a non-linear and household-specific relationshipbetween energy consumption and internaltemperature.

Measurement of behavioural responsesrequires thermostat settings to bemonitored directly, which is neitherstraightforward nor common. Instead,most studies monitor internaltemperatures and/or energy consumptionand focus upon temperature take-backand/or shortfall, which are frequentlymore relevant to public policy.Unfortunately, different studies using

different terms for the above concepts aswell as the same term for differentconcepts. The UK government uses theterm ‘comfort factor’ to refer to theshortfall from household insulationimprovements, but it could be misleadingto equate this with the direct reboundeffect.

The results from evaluation studies ofhousehold heating are difficult to compareowing to different approaches,inconsistent methodologies and differentdefinitions of both the dependent andindependent variables. Some studiesreport temperature take-back, somereport shortfall and some make anestimate of behavioural change.

Without exception, the studiesdemonstrate a clear shortfall in energysavings due to space heating efficiencymeasures. The absolute magnitude of thisdifference varies between studies, but isgenerally between 10% and 50% ofexpected savings. In some studies alarger shortfall was found – for example,an evaluation of a UK programme for lowincome households found that theintroduction of a more efficient heatingsystem had no impact on reducing fuelconsumption (Hong, et al., 2006).However, overall percent shortfall is highlycontingent on the accuracy of theengineering models, and attempts tocalibrate these models to specifichousehold conditions generally result in alower shortfall. The studies use a widerange of variables to explain shortfall, butonly initial energy consumption and theage of the home consistently influence the

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

24

14 For example, Hong et al. (2006) report the results of thermal camera imaging of houses following insulation improvements,which showed that 20% of the cavity wall area was missing insulation as well as 13% of the loft area.

S4134 12/10/07 13:56 Page 24

Measurement difficulties make estimation of direct rebound effects problematic at best andimpossible at worst. For many energy services, the relevant data is simply unavailable, while forothers the data must either be estimated or is subject to considerable error.

Measurement of useful work is extremely problematic, which partly explains why the empiricalliterature is largely confined to passenger transport, household heating and household coolingwhere suitable proxies are available. Aggregate measures of useful work are only available for asubset of energy services (e.g. vehicle kilometres for passenger transport), while disaggregatemeasures may require expensive monitoring of individual firms or households. In the case ofhousehold heating, a number of studies have monitored thermostat settings or internaltemperatures for individual households (Greening and Greene, 1998). But to assess the fullbenefits of the efficiency improvement, the temperature change in each room of the buildingwould need to be measured, while controlling for outside temperature, changes in occupancy andother factors. In addition, internal temperature is only one of the determinants of thermal comfort,with others including activity levels, air velocity, relative humidity and the mean radianttemperature of the surrounding surfaces (Fanger, 1970; Frey and Labay, 1988). Failure to takethese into account could lead to biased estimates of the direct rebound effect (Friedman, 1987;Greening and Greene, 1998).

Data on the energy consumption for individual consumer energy services is rarely available,except when those services are sub-metered as part of an evaluation exercise. Data on totalhousehold energy consumption may be relatively accurate, but techniques such as ‘conditionaldemand analysis’ are required to estimate the proportion going to lighting, water heating, cookingand so on (Parti and Parti, 1980). Problems may even arise in well-studied areas such as personalautomotive transport, where there is uncertainty over the proportion of total petrol and dieselconsumption attributable to passenger cars, as well as the distance travelled by different types ofvehicle (Schipper, et al., 1993).

If data on both energy consumption and useful work is available, system-wide energy efficienciescan be estimated. Alternatively, the demand for useful work could be estimated from data onenergy consumption and system-wide energy efficiency. But the latter is difficult to obtain, subjectto inaccuracy and (depending upon the system boundary) frequently dependent upon variablesother than the thermodynamic efficiency of conversion devices.

Empirical studies are more informative if the demand for useful work can be decomposed into theproduct of the number, capacity and utilisation of the relevant energy conversion devices, butusually this data is unavailable or subject to error. For example, several studies of personalautomotive transport combine data on the number of vehicles, the distance driven, total energyconsumption and average fuel efficiency that have not been independently and accuratelydetermined. (Schipper, et al., 1993). Also, the available data on useful work may not reflect thefull range of direct rebound effects.

Measurement difficulties also apply to exogenous variables that affect the demand for energy,useful work or energy efficiency, such as demographic and geographical factors. Omission of suchvariables could lead to bias if they are correlated with the dependent or independent variables,while inclusion of such variables could lead to error if they are measured or estimated inaccurately.

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

25

Box 3.1 Measurement issues for the direct rebound effect

S4134 12/10/07 13:56 Page 25

extent to which predicted savings arelikely to be achieved.

Some positive temperature take-back wasobserved in most studies; however it wasfrequently small and not alwaysstatistically significant. Take-backappeared to average between ~0.4ºC and0.8ºC. Of this, approximately half wasestimated to be accounted for by thephysical characteristics of the house andthe remainder by behavioural change. Inhomes with higher temperature take-back, the physical contribution remainsthe same and the behavioural contributionincreases. This take-back is not trivial: a1ºC increase in internal temperature mayincrease the energy consumption forspace heating by 10% or more.

Only a subset of the studies that measuretemperature take-back provide estimatesof the effect on energy savings and thosethat do vary widely, partly as aconsequence of the different metrics usedto present the effect. The likelihood ofobserving temperature take-back appearsto be negatively correlated with income,as well as internal householdtemperatures prior to the efficiencymeasure. However, these two explanatoryvariables are likely to be correlated. Aspre-intervention room temperaturesapproach 21ºC the magnitudes oftemperature take-back decreases owingto saturation effects.

Estimates of temperature take-backappear comparable between US and UKstudies, with most showing a relativelysmall temperature increase for theaverage home and higher temperatureincreases for homes with low initialtemperatures. UK average indoor

temperatures appear to be lower than inthe US, but have been increasing overrecent years irrespective of the energyefficiency measures installed.

In summary, the evaluation studiessuggest that standard engineering modelsmay overestimate the energy savingsfrom energy efficiency improvements inhousehold heating systems by up to onehalf – and potentially by more than thisfor low income households. However,temperature take-back only accounts fora portion of this shortfall and behaviouralchange only accounts for a portion of thetake-back. This suggests that the directrebound effect for this energy serviceshould typically be less than 30%.Rebound effects may be expected todecrease over time as average internaltemperatures increase.

3.3 Estimating directrebound effects fromeconometric studiesThe ‘evaluation’ approach to estimatingthe direct rebound effect requires data tobe collected on the demand for energy oruseful work both before and after anenergy efficiency improvement. This‘program focused’ approach is relativelyuncommon. Instead, most studies relyupon secondary data sources, frequentlycollected for other purposes that includeinformation on the demand for energy,useful work and/or energy efficiency. Thisdata can take a number of forms (e.g.cross-sectional, time-series) and apply todifferent levels of aggregation (e.g.household, region, country). Such studies

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

26

S4134 12/10/07 13:56 Page 26

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

27

typically use a variety of econometrictechniques to estimate elasticities,meaning the percentage change in onevariable following a percentage change inanother, holding other variables constant.If time-series data is available, anestimate can be made of short runelasticities, where the stock of conversiondevices is assumed to be fixed, as well aslong-run elasticities where it is variable.Cross-sectional data is usually assumed toprovide estimates of long-run elasticities.

Depending upon data availability, thedirect rebound effect may be estimatedfrom one of two energy efficiencyelasticities:15

E1 the elasticity of the demand forenergy with respect to energyefficiency

E2 the elasticity of the demand for usefulwork with respect to energy efficiency

(E2) is generally taken as a directmeasure of the rebound effect. Undercertain assumptions, it can be shownthat: (E1)=(E2)-1 (see Technical Report2). The actual saving in energyconsumption will only equal the predictedsaving from engineering calculationswhen the demand for useful work remainsunchanged following an energy efficiencyimprovement (i.e. (E2)=0).15a

But instead of using (E1) or (E2), themajority of studies estimate the reboundeffect from one of three price elasticities:

E3 the elasticity of the demand for usefulwork with respect to the price ofuseful work

E4 the elasticity of the demand for usefulwork with respect to the price ofenergy

E5 the elasticity of the demand forenergy with respect to the price ofenergy

Under certain assumptions, the negativeof (E3), (E4) or (E5) can be taken as anapproximation to (E2). The use of priceelasticities in this way implicitly equatesthe direct rebound effect to thebehavioural responses identified inSection 3.2 and ignores the other reasonswhy the demand for useful work maychange following an improvement inenergy efficiency.

Estimates of (E1) (E2) and (E3) requiredata on energy efficiency for the relevantenergy service, while estimates of (E3)(E4) and (E5) require data on energyprices. Generally, the latter tends to beboth more available and more accuratethan the former. Similarly, estimates of(E2) (E3) and (E4) require data on thedemand for useful work, while estimatesof (E1) and (E5) require data on thedemand for energy. Again, the lattertends to be both more available and moreaccurate than the former.

In principle, estimates of either (E1) or(E2) should provide the best

15 The rationale for the use of these elasticities, and the relationship between them, is explained in detail in Technical Report 2.In the case of personal automotive transport, they could correspond to: (E1) the elasticity of the demand for motor-fuel (forpassenger cars) with respect to kilometres per litre (E2) the elasticity of the demand for vehicle kilometres with respect tokilometres per litre; (E3) the elasticity of the demand for vehicle kilometres with respect to the cost per kilometre; (E4) theelasticity of the demand for vehicle kilometres with respect to the price of motor-fuel; and (E5) the elasticity of the demand formotor-fuel with respect to the price of motor-fuel.

15a Under these circumstances, (E1)=-1. A positive rebound effect implies that (E2)>0 and 0>(E1)>-1, while backfire impliesthat (E2)>1 and (E1)>0.

S4134 12/10/07 13:56 Page 27

approximation to the direct reboundeffect. However, most data sets provideonly limited variation in energy efficiency,with the result that estimates of (E1) and(E2) have a large variance. This is anotherreason why estimates of (E1) and (E2) arerelatively rare.

In contrast, (E3) generally providessignificantly greater variation in theindependent variable. This is because theprice of useful work depends upon theratio of energy prices to energy efficiencyand most data sets include considerablecross-sectional or longitudinal variation inenergy prices. In principle, rationalconsumers should respond in the sameway to a decrease in energy prices as theydo to an improvement in energy efficiency(and vice versa), since these should havean identical effect on the price of usefulwork. However, there may be a number ofreasons why this ‘symmetry’ assumptiondoes not hold (see below). If so,estimates of the direct rebound effect thatare based upon (E3) could be biased.

In many cases, data on energy efficiencyis either unavailable or inaccurate. Inthese circumstances, (E4) and (E5) allowthe direct rebound effect to be estimatedsolely from data on energy prices. Butthese are only valid if: first, consumersrespond in the same way to a decrease inenergy prices as they do to animprovement in energy efficiency (andvice versa); and second, energy efficiencyis unaffected by changes in energy prices.Both these assumptions are likely to be

flawed, but the extent to which this leadsto biased estimates of the direct reboundeffect may vary widely from one energyservice to another and between the shortand long term.

Under certain assumptions, the own-priceelasticity of energy demand (E5) can beshown to provide an upper bound for thedirect rebound effect (see TechnicalReport 2). As a result, the voluminousliterature on energy price elasticities maybe used to place some bounds on thelikely magnitude of the direct reboundeffect for different energy services indifferent sectors. This was the approachtaken by Khazzoom (1980), who pointedto evidence that the long-run own-priceelasticity of energy demand for waterheating, space heating and cookingexceeded (minus) unity in somecircumstances, implying that energyefficiency improvements for theseservices could lead to backfire (Taylor, etal., 1977). However, reviews of thisliterature generally suggest that energydemand is inelastic in the majority ofsectors in OECD countries (i.e. ) (Dahl andSterner, 1991; Dahl, 1993; 1994; Espey,1998; Graham and Glaister, 2002; Hanley,et al., 2002; Espey and Espey, 2004).

As an illustration, the upper bound for thedirect rebound effect for personalautomotive transport can be estimatedfrom Hanley et al’s (2002) meta-analysis16 of 51 empirical estimates of thelong-run own-price elasticity of motor-fueldemand (E5). Taking the mean values,

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

28

16 A meta-analysis combines the results of several empirical studies that provide comparable quantitative estimates of aparticular variable, such as the own-price elasticity of energy demand for a particular sector. In energy economics, a commonapproach is to use the estimated elasticities as the dependent variable in a multiple regression analysis that seeks to explain thevariation in the results in terms of factors such as model structure (Bergh and Button, 1999).

S4134 12/10/07 13:56 Page 28

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

29

this suggests an upper bound for the long-run direct rebound effect for this energyservice of 64%.17 For comparison, threeestimates of the elasticity of distancetravelled by private cars with respect tothe price of motor-fuel (E4) suggest asmaller long-run direct rebound effect ofonly 30% (Hanley, et al., 2002). Thedifference between the two suggests thatfuel prices have a significant influence onvehicle fuel efficiency over the long termand shows the extent to which the use of(E5) may lead to the direct rebound effectbeing overestimated.

For the purpose of estimating reboundeffects, own-price elasticities for energydemand are most useful when thedemand relates to a single energy service,such as refrigeration. They are less usefulwhen (as is more common) the measureddemand derives from a collection ofenergy services, such household fuel orelectricity consumption. In this case, alarge own-price for fuel or electricitydemand may suggest that improvementsin the ‘overall’ efficiency of fuel orelectricity use by the household will leadto large direct rebound effects (and viceversa). It may also suggest that the directrebound effect for the energy servicesthat dominate fuel or electricityconsumption may be large. This isrelevant for household heating, forexample, since this typically accounts formore than two thirds of household fuelconsumption. But many households use

more than one energy carrier for heatingand there may also be scope for long-termsubstitution between different energycarriers. In these circumstances the own-price elasticity for fuel consumption couldoverestimate the own-price elasticity offuel use for heating. At the same time,such aggregate data could underestimatethe own-price elasticity of energy demandfor other energy services. For example, itis possible that the energy demand forrefrigeration is elastic – suggesting thepossibility of large direct rebound effects -but if overall household electricity demandis inelastic, such a possibility is disguised.

Estimates of the own-price elasticity ofhousehold energy demand vary with theenergy carrier, the type of household, theparticular country and/or region, the levelof household income18 and themethodological approach. For electricity,the results of a meta-analysis of morethan 120 estimates suggests an upperbound for the short-term rebound effectfor all household electricity servicescombined of 20-35% and an upper boundfor the long-term effect of 80-85% (Espeyand Espey, 2004). Evidence on fuel priceelasticities is more variable and lessrobust, but generally suggests a lowerprice elasticity than for electricity (Bakerand Blundell, 1991; Ferrer-i-Carbonell, etal., 2002). Overall, and contrary toKhazzoom (1980), the evidence suggeststhat own-price elasticities of householdfuel and electricity demand are less than

17 However, the mean values disguise a large variance in the results and using the full range of estimates suggests an upperbound of between 0 and 181%. This partly reflects the wide range of countries and time period on which the estimates are based.The corresponding mean value for the short run is 25%.

18 Price elasticities are frequently found to be higher among low income groups. For example, Baker et al. (1989) found that,conditional upon appliance ownership, the own-price elasticity of gas consumption by UK households was two times larger forthe lowest income decile than for top income decile. However, Nesbakken (1999) found that the price elasticity of Norwegianhousehold energy demand increases with income. Also, low income households often live in rented accommodation and hencehave less scope for changing equipment such as heating systems over the long term (Poyer and Williams, 1992).

S4134 12/10/07 13:56 Page 29

unity and hence that energy efficiencyimprovements are unlikely to lead tobackfire (at least, not from direct reboundeffects alone).

Whatever their origin, estimates of priceelasticities should be treated with greatcaution. Aside from the difficulties ofestimation, behavioural responses arecontingent upon technical, institutional,policy and demographic factors that varywidely between different groups and overtime. Demand responses are known tovary with the level of prices, the origin ofprice changes (e.g. exogenous versuspolicy induced), expectations of futureprices, government fiscal policy (e.g.recycling of carbon tax revenues),saturation effects and other factors(Barker, et al., 1995). The past is notnecessarily a good guide to the future inthis area, and it is possible that the verylong-run response to price changes mayexceed those found in empirical studiesthat rely upon data from relatively shorttime periods.

Instead of relying upon contestedestimates of (E5), more accurateinformation on the magnitude of directrebound effects may be obtained fromstudies that directly estimate (E3).However, data is only available onrelevant measures of useful work for asubset of energy services. As aconsequence, the reliable evidence basefor the direct rebound effect is largelyfocused on personal automotivetransportation, household heating andspace cooling, where proxy measures ofuseful work are most readily available.These energy services form a significantcomponent of household energy demandin OECD countries and may be expected

to be relatively price elastic. Econometricevidence for other consumer energyservices is very limited, while that forproducers is almost non-existent. Theevidence base also exhibits a notablegeographical bias, with the majority ofstudies referring to the United States andwith very few studies of energy services indeveloping countries.

Our review of the empirical literature inthis area is therefore largely confined tostudies that estimate (E1), (E2) or (E3).These vary widely in terms of type ofdata, model structure, functional form andestimation techniques, which complicatesthe comparison of results. Given thisdiversity and the limited number ofstudies available we have not conducted aformal meta-analysis, but have insteadsummarised key results and identifiedpossible sources of error. The result of thissurvey are summarised in the next twosections.

3.4 Direct rebound effectsfor private automotivetransport By far the best studied area for the directrebound effect is personal transport byprivate car. Most studies refer to the US,which is important since fuel prices, fuelefficiencies and residential densities arelower than in Europe, car ownership levelsare higher and there is less scope forswitching to alternative transport modes.

Studies estimating (E1), (E2) or (E3) varyconsiderably in terms of the data usedand specifications employed. Most studiesuse aggregate data which can capture

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

30

S4134 12/10/07 13:56 Page 30

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

31

long-term effects on demand such as fuelefficiency standards, while householdsurvey data can better describe individualbehaviour at the micro level. Aggregatestudies face numerous measurementdifficulties, however, while disaggregatestudies produce results that are moredifficult to generalise. The relevantmeasure for useful work varies betweentotal distance travelled, distance travelledper capita, distance travelled per licenseddriver, distance travelled per householdand distance travelled per vehicle.

Studies using aggregate time-series andcross-sectional data estimate the long-rundirect rebound effect for personalautomotive transport to be somewherebetween 5% and 30%. While there isdisagreement over the appropriatespecification the limited number of datapoints available makes it difficult to settlethe issue from this type of data alone(Greene, 1992; Jones, 1993).19

Aggregate panel data20 provides a morerobust basis for estimates of the directrebound effect, owing to the greaternumber of observations. Johansson andSchipper’s (1997) cross-country studygives a best guess for the long-run directrebound effect of 30%, while bothHaughton and Sakar (1996) and Smalland van Dender (2005) converge on along-run value of 22% for the US. Small

and van Dender’s study incorporatessome important methodologicalinnovations and is one of the few to testthe hypothesis that the direct reboundeffect declines with income (Box 3.2).

Studies using household survey datasources provide less consistent estimatesof the direct rebound effect for personalautomotive transport and several of theseestimates are higher than those fromaggregate data sources. Three US studiesuse data from the same source butproduce estimates of the direct reboundeffect that range from 0% to 87%(Goldberg, 1996; Puller and Greening,1999; West, 2004).21 This diversitysuggests that the results fromdisaggregate studies should beinterpreted with more caution. One of themost rigorous studies using disaggregatedata is by Greene et al. (1999), whoestimate the US average long-run directrebound effect to be 23% - consistentwith the results of studies using aggregatedata.

Taken together, our review of 17 studiessuggests that the long-run direct reboundeffect for personal automotive transportlies somewhere between 10% and 30%.The relative consensus on estimates,despite wide differences in data andmethodologies suggests that the findingsare relatively robust. Moreover, most of

19 The dispute relates to the appropriate treatment of serial correlation and lagged dependent variables. Serial correlation meansthat the error in one time period is correlated with the error from one or more previous time periods, perhaps as a result of theinfluence of unobserved variables that persist over time. Identifying and correcting for serial correlation is a major issue in time-series econometrics and is relevant to many studies of the direct rebound effect.

20 A pooled cross-section is a cross-sectional sample from a population, taken at two or more intervals in time. Panel data issimilar, but with data from the same units in each sample period.An example would be data on motor-fuel consumption andrelated variables from each US state over a period 1972 to 2000.

21 All use the US Consumer Expenditure Survey, although covering different time periods and supplemented by differing sourcesof information on vehicle fuel efficiency. The figure of 87% is from West (2004) and is likely to be an overestimate of the directrebound effect since it derives from the estimated elasticity of distance travelled with respect to operating costs, which includesmaintenance and tyre costs.

S4134 12/10/07 13:56 Page 31

these studies assume that the response toa change in fuel prices is equal in size tothe response to a change in fuel efficiency,but opposite in sign. Few studies test thisassumption explicitly and those that doare either unable to reject the hypothesisthat the two elasticities are equal, or findthat the fuel efficiency elasticity is lessthan the fuel cost per kilometer elasticity.The implication is that the direct reboundeffect may lie towards the lower end ofthe above range.

The extent to which the direct reboundeffect for personal automotive traveldeclines with income remains unclear,although the methodologically rigorousstudies by Small and van Dender (2005)and Greene et al. (1999a) both suggestthat it does. Measurement problemsremain an issue for aggregate studies(Schipper, et al., 1993), as does thegeographical bias towards the UnitedStates. The available evidence isinsufficient to determine whether directrebound effects are larger or smaller inEurope, but it is notable that the meta-analysis by Espey (1998) found nosignificant difference in long-run own-price elasticities of gasoline demand.Overall, it must be concluded that directrebound effects in this sector have notobviated the benefits of technicalimprovements in vehicle fuel efficiency.Between 70% to 100% of the potentialbenefits of such improvements appear tohave been realised in reducedconsumption of motor-fuels.

3.5 Direct rebound effectsfor other consumer energyservicesThe next best studied area for directrebound effects is household heating,although relatively few studies estimate(E1), (E2) or (E3) for this energy serviceand even fewer investigate reboundeffects specifically. The available studiesrely upon detailed household survey dataand exhibit even greater diversity in termsof the variables measured and themethodologies adopted. Four of the mostrigorous studies are briefly describedhere.

Schwarz and Taylor (1995) use cross-sectional data from 1188 single family UShouseholds, including measurements ofthermostat settings. They estimate anequation for the thermostat setting as afunction of energy prices, externaltemperature, heated area, householdincome and an engineering estimate ofthe thermal resistance of the house. Theirdata also allows them to estimate thedemand for useful work for space heating.On the basis of thermostat settings, theirestimate of (E2) suggests a long-rundirect rebound effect of 0.6% to 2.0%,while on the basis of the demand foruseful work their estimate suggests alarger effect of 1.4% to 3.4%. Whilethese estimates are smaller than thosefrom other studies, the differencebetween the two is comparable to thatbetween behavioural responses andtemperature take-back identified by theevaluation studies.

Hseuh and Gerner (1993) use comparabledata from 1281 single family detached

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

32

S4134 12/10/07 13:56 Page 32

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

33

Small and van Dender (2005) provide one of the most methodologically rigorousestimates of the direct rebound effect for personal automotive transport. They estimatean econometric model explaining the amount of travel by passenger cars as a functionof the cost per mile and other variables. By employing simultaneous equations forvehicle numbers, average fuel efficiency and vehicle miles travelled, they are able allowfor the fact that fuel efficiency is endogenous: i.e. more fuel-efficient cars mayencourage more driving, while the expectation of more driving may encourage thepurchase of more fuel-efficient cars. Their results show that failing to allow for this canlead the direct rebound effect to be substantially overestimated.

Small and Van Dender use aggregate data on vehicle numbers, fuel efficiency, gasolineconsumption, vehicle miles travelled and other variables for 50 US states and theDistrict of Columbia covering the period 1961 to 2001. This approach providesconsiderably more observations than conventional aggregate time-series data, while atthe same time providing more information on effects that are of interest to policymakersthan do studies using household survey data. The effect of the CAFE standards onvehicle fuel efficiency are estimated by incorporating a variable representing the gapbetween the fuel efficiency standard and an estimate of the efficiency that would havebeen chosen in the absence of the standards, giving prevailing fuel prices.

Small and Van Dender estimate the short-run direct rebound effect for the US as a wholeto be 4.5% and the long-run effect to be 22%. The former is lower than most of theestimates in the literature, while the latter is close to the consensus. However, theyestimate that a 10% increase in income reduces the short-run direct rebound effect by0.58%. Using US average values of income, urbanisation and fuel prices over the period1997-2001, they find a direct rebound effect of only 2.2% in the short-term and 10.7%in the long-term - approximately half the values estimated from the full data set. If thisresult is robust, it has some important implications. However, two-fifths of the estimatedreduction in the rebound effect derives from the assumption that the magnitude of thiseffect depends upon the absolute level of fuel costs per kilometre. But since the relevantcoefficient is not statistically significant, this claim is questionable.

Although methodologically sophisticated, the study is not without its problems. Despitecovering 50 states over a period of 36 years, the data provides relatively little variationin vehicle fuel efficiency making it difficult to determine its effect separately from thatof fuel prices. Direct estimates of (E2) are small and statistically insignificant, whichcould be interpreted as implying that the direct rebound effect is approximately zero,but since this specification performs rather poorly overall, estimates based upon (E3)are preferred. Also, the model leads to the unlikely result that the direct rebound effectis negative some states (Harrison, et al., 2005). This raises questions about the use ofthe model for projecting declining rebound effects in the future, since increasingincomes could make the estimated direct rebound effect negative in many states.

Box 3.2 The declining direct rebound effect

S4134 12/10/07 13:56 Page 33

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

34

households in the US, dating back to1981. Their dataset includescomprehensive information on applianceownership and demographiccharacteristics, which allows them tocombine econometric and engineeringmodels to estimate the energy use forspace heating. On the basis of an estimateof (E5), conditional on the existing level ofenergy efficiency, the short-run directrebound effect can be estimated as 35%for electrically heated homes and 58% forgas heated homes.

Klein (1987; 1988) uses comparable datafrom more than 2000 US households,supplemented with an engineering modelof the thermal performance of buildingsand data on the capital cost of equipment.This allows the estimation of asophisticated ‘household production’model, involving simultaneous equationsfor the total cost of space heat, the shareof energy in the total cost of space heatand the demand for space heat. On thebasis of (E3), Klein’s model suggests ashort-run direct rebound effect in therange 25% to 29%.22

Guertin et al. (2003) use detailed surveydata from 440 single family Canadianhouseholds and apply ‘frontier analysis’ toestimate the energy efficiency of therelevant appliances. This in turn allowsthe energy consumption for spaceheating, water heating andappliances/lighting to be estimated. Onthe basis of (E3), Guertin et al’s modelsuggests a long-run mean rebound effect

of 38%, varying from 29% for highincome groups to 47% for low incomegroups.23

Other estimates of direct rebound effectsfor household heating include Haas andBiermayer (2000) who use a variety oftechniques to reach an estimate of 15-30% for German households, and Douthitt(1986) whose study of 370 Canadianhouseholds suggest short-run effects inthe range 10-17% and long-run effects inthe range 25-60%.

Reliable estimates of the direct reboundeffects for household heating thereforerange from 10% to 58% in the short-runand 1.4% to 60% in the long-run, with asuggestion that rebound effects are largerfor low income groups. As with theevaluation studies, the definition of thedirect rebound effect is not consistentbetween the above studies, the responsevaries between different households andthe results from one time period andgeographical area may not translate toother circumstances. Nevertheless, for thepurpose of policy evaluation, a figure of30% for the direct rebound effect forhousehold heating would appear areasonable assumption.

For space cooling in households, twostudies provide estimates of directrebound effects that are within the rangeof those reported for space heating (i.e.1-26%) (Hausman, 1979; Dubin, et al.,1986). However, these relatively oldstudies use small sample sizes and wereconducted during period of rising energy

22 Greening et al’ (1998) quote a figure of 40% from the study, but this overestimates the direct rebound effect since it is basedupon the elasticity of heat demand with respect to the generalised cost of space heat, which includes capital and maintenancecosts.

23 The variation in (E3) between income groups was much greater for space heating than for other energy services.

S4134 12/10/07 13:56 Page 34

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

35

prices. Their results may not betransferable to many European countries,given the differences in house types andclimatological conditions. Also, bothstudies focus solely upon changes inequipment utilisation. To the extent thatownership of space cooling technology israpidly increasing in many countries,demand from ‘marginal consumers’ maybe an important consideration, together

with increases in system capacity amongexisting users.

Evidence for water heating is also verylimited, although Guertin et al. (2003)provides estimates in the range 34 to38%, which is greater than the resultsfrom US evaluation studies reported byNadel (1993). A methodologically rigorousstudy of direct rebound effects for clotheswashing suggests that direct rebound

Davis (2007) provides a unique example of an estimate of direct rebound effects forhousehold clothes washing - which together with clothes drying accounts for around onetenth of household energy consumption. The estimate is based upon a US government-sponsored field trial of high-efficiency washing machines involving 98 participants.These machines use 48% less energy per wash than standard machines and 41% lesswater.

While participation in the trial was voluntary, both the utilisation of existing machinesand the associated consumption of energy and water was monitored for a period of twomonths prior to the installation of the new machine. This allowed household specificvariations in utilisation patterns to be controlled for and permitted unbiased estimatesto be made of the price elasticity of machine utilisation.

The monitoring allowed the marginal cost of clothes washing for each household to beestimated. This was then used as the primary independent variable in an equation forthe demand for clean clothes in kg/day (useful work). Davis found that the demand forclean clothes increased by 5.6% after receiving the new washers, largely as a result ofincreases in the weight of clothes washed per cycle rather than the number of cycles.While this could be used as an estimate of the direct rebound effect, it results in partfrom savings in water and detergent costs. If the estimate was based solely on thesavings in energy costs, the estimated effect would be smaller. This suggests that onlya small portion of the gains from energy efficient washing machines will be offset byincreased utilisation.

Davis estimates that time costs form 80-90% of the total cost of washing clothes. Theresults therefore support the theoretical prediction that, for time intensive activities,even relatively large changes in energy efficiency should have little impact on demand.Similar conclusions should therefore apply to other time-intensive energy services thatare both produced and consumed by households, including those provided bydishwashers, vacuum cleaners, televisions, power tools, computers and printers.

Box 3.3 Direct rebound effects for clothes washing

S4134 12/10/07 13:56 Page 35

effects for ‘minor’ energy services shouldbe relatively small (Box 3.3). However,this confines attention to households thatalready have automatic washing machinesand therefore excludes rebound effectsfrom marginal consumers.

Table 3.1 summarises the results of oursurvey of econometric estimates of thedirect rebound effect. Despite themethodological diversity, the results forindividual energy services are broadlycomparable. This suggests that theevidence for direct rebound effects isrelatively robust to different datasets andmethodologies.

These relatively modest estimates fordirect rebound effects are based uponstudies in OECD countries and are unlikelyto be representative of conditions indeveloping countries. Hence, two studiesfrom developing countries deserve amention. In a rare study of traditionalfuels, Zein-Elabdin (1997) estimatesdirect rebound effects from fuel-efficientstoves in Khartoum by multiplying the fuelefficiency improvement by the product ofthe estimated income elasticity of demandfor charcoal and the share of charcoal inthe household budget. Allowing for the

additional (indirect) effects on the price ofcharcoal leads to an estimated reboundeffect of 42%.

Roy (2000) reports the results of agovernment programme that freelydistributed solar-charged battery lamps toa rural village in India. These lampsprovided significantly better lightingquality than the kerosene lamps theydisplaced. Daily hours of lightingincreased from two to four, provided by acombination of new and old lamps sincethe solar lamps were limited in operatinghours. This led to a direct rebound effectof approximately 50% (80% for somehouseholds). Moreover, since the ‘saved’kerosene was either used for cooking orsold, accounting for indirect effectssuggests that the programme achieved noenergy savings - although significantbenefits to welfare. Since the lamps werefree and kerosene subsidised, this may bean atypical example. However, theprimary source of the large rebound effectwas the large unsatisfied demand forlighting. This condition applies to a rangeof energy services in developing countriesand especially for the 1.6 billionhouseholds who currently lack access to

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

36

End-Use Range of ‘Best guess’ No. of Studies Degree Values in of Confidence

Evidence Base

Personal automotive 5-87% 10-30% 17 Hightransport

Space heating 1.4-60% 10-30% 9 Medium

Space cooling 1-26% 1-26% 2 Low

Other consumer 0-49% <20% 3 Lowenergy services

Table 3.1 Estimates of the long-run direct rebound effect for consumer energyservices in the OECD

S4134 12/10/07 13:56 Page 36

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

37

electricity and the 2.5 billion who relyupon biomass for cooking.

3.6 Sources of bias inestimates of the directrebound effectThe use of price elasticities to estimatethe direct rebound effect creates the riskof bias. Most studies assume that changesin energy prices have an opposite effect tocomparable changes in energy efficiencyand that any changes in energy efficiencyderive solely from outside the model (i.e.energy efficiency is ‘exogenous’). Inpractice, both of these assumptions maybe incorrect.

First, while changes in energy prices aregenerally not correlated with changes inother input costs, changes in energyefficiency may be. In particular, higherenergy efficiency may only be achievedthrough the purchase of new equipmentwith higher capital costs than less efficientmodels. Hence, estimates of the directrebound effect that rely primarily uponhistorical and/or cross-sectional variationsin energy prices could overestimate thedirect rebound effect, since the additionalcapital costs required to improve energyefficiency will not be taken into account(Henly, et al., 1988).

Second, energy price elasticities tend tobe higher for periods with rising pricesthan for those with falling prices (Gately,1992; 1993; Dargay and Gately, 1994;1995; Haas and Schipper, 1998). Forexample, Dargay (1992) found that thereduction in UK energy demand followingthe price rises of the late 1970s was five

times greater than the increase indemand following the price collapse of themid-1980s. An explanation may be thathigher energy prices induce technologicalimprovements in energy efficiency, whichmay also become embodied in regulations(Grubb, 1995). Also, investment inmeasures such as thermal insulation islargely irreversible over the short tomedium-term. But the appropriate proxyfor improvements in energy efficiency isreductions in energy prices. Since manystudies based upon time series dataincorporate periods of rising energyprices, the estimated price elasticitiesmay overestimate the response to fallingenergy prices. As a result, such studiescould overestimate the direct reboundeffect.

Third, while improved energy efficiencymay increase the demand for useful work(e.g. you could drive further afterpurchasing an energy-efficient car), it isalso possible that the anticipated highdemand for useful work may increase thedemand for energy efficiency (e.g. youpurchase an energy-efficient car becauseyou expect to drive further). In thesecircumstances, the demand for usefulwork depends on the price of useful work,which depends upon energy efficiency,which depends upon the demand foruseful work (Small and Van Dender,2005). Hence, the direct rebound effectwould not be the only explanation for anymeasured correlation between energyefficiency and the demand for usefulwork. This so-called ‘endogeneity’ can beaddressed through the use ofsimultaneous equation models, but theseare relatively uncommon owing to theirgreater data requirements. If, instead,

S4134 12/10/07 13:56 Page 37

studies include the ‘endogenous’variable(s) within a single equation and donot use appropriate techniques toestimate this equation, the resultingestimates could be biased. Several studiesof direct rebound effects could be flawedfor this reason.

Finally, consumers may be expected totake the full costs of energy services intoaccount when making decisions about theconsumption of those services and theseinclude the time costs associated withproducing and/or using the relevantservice - for example, the time required totravel from A to B. Indeed, the increase inenergy consumption in industrial societiesover the past century may have beendriven in part by attempts to ‘save time’(and hence time costs) through the use oftechnologies that allow tasks to becompleted faster at the expense of usingmore energy. For example, travel byprivate car has replaced walking, cyclingand public transport; automatic washingmachines have replaced washing by hand;fast food and ready meals have replacedtraditional cooking and so on. While not allenergy services involve such trade-offs,many do (compare rail and air travel forexample). Time costs may beapproximated by hourly wage rates andsince these have risen more rapidly thanenergy prices throughout the last century,there has been a strong incentive tosubstitute energy for time (Becker, 1965).If time costs continue to increase inimportance relative to energy costs, thedirect rebound effect for many energyservices should become less important –

since improvements in energy efficiencywill have an increasingly small impact onthe total cost of useful work (Binswanger,2001). This suggests that estimates of thedirect rebound effect that do not controlfor increases in time costs (which iscorrelated with increases in income) couldpotentially overestimate the directrebound effect. Box 3.2 shows how thiscould be particular relevant to directrebound effects in transport.

The consideration of time costs also pointsto an important but relatively unexploredissue: increasing time efficiency may leadto a parallel ‘rebound effect with respectto time’ (Binswanger, 2001; Jalas, 2002).For example, faster modes of transportmay encourage longer commutingdistances, with the time spent commutingremaining broadly unchanged.24 So insome circumstances energy consumptionmay be increased, first, by trading offenergy efficiency for time efficiency (e.g.choosing air travel rather than rail) andsecond, by the rebound effects withrespect to time (e.g. choosing to travelfurther).

3.7 Summary• Evidence for the direct rebound effect

for automotive transport andhousehold heating within developedcountries is relatively robust.Evidence for direct rebound effects forother consumer energy services ismuch weaker, as is that for energyefficiency improvements by

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

38

24 In the UK, the time spent travelling increased by only 17% between 1975 in 2005, while the total distance travelled increasedby 52% (Department of Transport, 2006).

S4134 12/10/07 13:56 Page 38

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

39

producers. Evidence is particularlyweak for energy efficiencyimprovements in developing countriesalthough theoretical considerationssuggest that direct rebound effects inthis context will be larger than thosein developed countries.

• Under certain assumptions, estimatesof the own-price elasticity of energydemand for an individual energyservice should provide an upperbound for the direct rebound effect forthat service. If the measured energydemand relates to a group of energyservices (e.g. household fueldemand), the own price elasticityshould provide an approximate upperbound for the weighted average ofdirect rebound effects for thoseservices. Since the demand for energyis generally found to be inelastic inOECD countries, the long-run directrebound effect for most energyservices should be less than 100%.

• For personal automotive transport,household heating and householdcooling in OECD countries, the meanvalue of the direct rebound effect islikely to be less than 30% and may becloser to 10% for transport. Moreover,the direct rebound effect is expectedto decline in the future as demandsaturates and income increases. Boththeoretical considerations and theavailable empirical evidence suggestthat direct rebound effects should besmaller for other consumer energy

services where energy forms a smallproportion of total costs. Hence, atleast for OECD countries, directrebound effects should only partiallyoffset the energy savings from energyefficiency improvements in consumerenergy services.

• These conclusions are subject to anumber of qualifications, including therelatively limited time periods overwhich direct rebound effects havebeen studied and the restrictivedefinitions of ‘useful work’ that havebeen employed. For example, currentstudies only measure the increase indistance driven for automotivetransport and do not measurechanges in vehicle size. Reboundeffects for space heating and otherenergy services are also higheramong low income groups and moststudies do not account for ‘marginalconsumers’ acquiring services such asspace cooling for the first time.

• The methodological quality of many‘evaluation studies’ is poor, while theestimates from many econometricstudies appear vulnerable to bias. Themost likely effect of the latter is tolead the direct rebound effect to beoverestimated. Considerable scopeexists for improving estimates of thedirect rebound effect for the energyservices studied here and forextending estimates to include otherenergy services.

S4134 12/10/07 13:56 Page 39

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

40

S4134 12/10/07 13:56 Page 40

15

20 25

30

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

41

4. The evidence for indirect and economy-wide rebound effects

This section summarises the empiricalevidence for indirect and economy-widerebound effects, focusing upon the limitednumber of studies that providequantitative estimates of these effects.This is separate from the more ‘indirect’evidence for economy-wide reboundeffects that is the subject of Section 5. Afull examination of this evidence iscontained in Technical Report 4 andTechnical Report 5.

Indirect rebound effects derive from twosources: the energy required to produceand install the measures that improveenergy efficiency, such as thermalinsulation, and the indirect energyconsumption that results from suchimprovements. The first of these relatesto energy consumption that occurs priorto the energy efficiency improvement,while the second relates to energyconsumption that follows theimprovement. Sections 4.1 and 4.2describe the mechanisms responsible foreach.

Section 4.3 examines the limitedempirical evidence for indirect reboundeffects, focusing on studies that estimatethe ‘embodied’ energy associated withdifferent categories of consumer goodsand services. Despite the apparentpotential of this approach, there appear tobe very few applications to the reboundeffect.

Section 4.4 examines the evidence foreconomy-wide rebound effects availablefrom computable general equilibrium(CGE) models of the macro-economy.These simulate the full range ofmechanisms responsible for reboundeffects, but the realism and policy

relevance of this approach is open toquestion. Also, despite the widespreaduse of such models, there are only ahandful of applications to the reboundeffect. Section 4.5 describes analternative ‘macro-econometric’ approachthat can overcome some of theseweaknesses, and summarises the resultsof a recent application of this approach tothe UK economy. Section 4.6 concludes.

4.1 Embodied energy – thelimits to substitutionMany improvements in energy efficiencycan be understood as the ‘substitution’ ofcapital for energy within a particularsystem boundary. For example, thermalinsulation (capital) may be substituted forfuel (energy) to maintain the internaltemperature of a building at a particularlevel. It is these possibilities that form thebasis of estimated ‘energy saving’potentials in different sectors. However,estimates of energy savings typicallyneglect the energy consumption that isrequired to produce and maintain therelevant capital - frequently referred to asembodied energy. For example, energy isrequired to produce and install homeinsulation materials and energy efficientmotors. Substituting capital for energytherefore shifts energy use from thesector in which it is used to sectors of theeconomy that produce that capital. As aresult, energy use may increaseelsewhere in the economy (Kaufmann andAzary-Lee, 1990).

Figure 4.1 provides a graphicalinterpretation of this process (Stern,1997). Here, the function F(K) represents

S4134 12/10/07 13:56 Page 41

15

20 25

30

the different combinations of energy (E)and capital (K) that may be used toprovide a given level of output for aparticular firm or sector. But this capitalalso has indirect energy consumptionassociated with it, elsewhere in theeconomy, represented by the functionG(K). The summation of the two gives thetotal energy H(K) used in the economy asa whole to produce the given level ofoutput. It can be seen that: first, the netenergy savings from the substitution ofcapital for energy will be less than theenergy savings within the firm or sector inwhich the substitution takes place; andsecond, when capital inputs exceed acertain level (K’), the indirect energyconsumption will exceed the direct energysavings – leading to backfire for theeconomy as a whole, even when theindividual firm or sector reduces energyconsumption and when output from thissector is unchanged.

In practice, backfire from this sourcealone appears rather unlikely, since thecost of an energy efficient technologyshould reflect the cost of the embodiedenergy (Webb and Pearce, 1975). If thelatter exceeds the saving in energy costs,it is unlikely that the investment would becost-effective.25 However, this assumesthat the sole benefit of the investment is

the reduced energy costs, which may notalways be the case. Also, marketimperfections may distort the relevantprices and costs.

In contrast to other sources of theeconomy-wide rebound effect, thecontribution from this source may beexpected to be smaller in the long-termthan in the short-term. This is because theembodied energy associated with capitalequipment is analogous to a capital costand hence diminishes in importancerelative to ongoing energy savings as thelifetime of the investment increases.

Some authors argue that similarconclusions apply to the substitution oflabour for energy, since energy is alsorequired to feed and house workers andthereby keep them economicallyproductive (Kaufmann, 1992). However,there is some dispute over whether andhow to account for the ‘energy cost oflabour’ (Costanza, 1980).26 Similarly,while economists conventionallydistinguish between substitution andtechnical change (Box 2.1), the latter isalso associated with indirect energyconsumption since it is embodied incapital goods and skilled workers (Sternand Cleveland, 2004).

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

42

15

20 25

30

25 Note that this argument applies to measures of energy consumption weighted by the relative price of different energy carriersand not to energy consumption measured simply in terms of heat content.

26 Costanza (1980) estimates the energy cost of labour as the energy associated with all personal consumption expenditures.These energy costs are then assigned to individual goods and services in proportion to the labour required to produce them.Double counting is avoided by changing the boundaries of the traditional economic input-output analysis. The net result is togreatly increase the embodied energy estimated to be associated with labour-intensive goods and services. With this approach,the ‘embodied energy intensity’ of most sectors (excluding primary energy) is found to be broadly comparable. However, thisconclusion depends entirely upon the particular methodology for calculating the energy cost of labour. This is quite different fromconventional accounting approaches, which estimate much lower energy intensities for many sectors and greater variationbetween them. It also implicitly assumes that all personal consumption expenditures are necessary to support labour, whichappears unjustified.

S4134 12/10/07 13:56 Page 42

15

20 25

30

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

43

In principle, an assessment of theembodied energy associated with aparticular energy efficiency improvementshould take into account the relevantalternatives. For example, a mandatoryrequirement to replace existingrefrigerators with more energy efficientmodels may either increase or decreaseaggregate energy consumption over aparticular period of time, depending uponthe age of the existing stock, the lifetimeof the new stock, and the direct andindirect energy consumption associatedwith different models of refrigerator. Inpractice, however, such estimates appearto be rare, with most analysts focusinginstead upon the ‘energy return on energyinvested’27 for different energy supplyoptions (Cleveland, 1992).28

4.2 Secondary effectsEnergy efficiency improvements maythemselves change the demand for othergoods and services. For example, thepurchase of a more fuel-efficient car mayreduce demand for public transport, but atthe same time increase the demand forleisure activities that can only be accessedwith a private car (Binswanger, 2001).Each of these goods and services will havean indirect energy consumptionassociated with them and the changedpattern of demand may either increaseoverall energy consumption or reduce it.The net impact of such effects will bespecific to both individual technologiesand individual consumers and can also bevery difficult to estimate.

Figure 4.1 The limits to substitution Source: Stern (1997)

Energy - E

Capital - K

H (K)

F (K)

G (K)

K’

27 The Energy Returned on Energy Invested (EroEI) is the ratio of the usable energy acquired from a particular energy resourceto the amount of energy expended to obtain that energy resource. In principle, when the EROEI of a resource is equal to orlower than 1, that energy source can no longer be used as a primary source of energy. However, this measure neglects therelative economic productivity of different energy forms. When this is taken into account, resources with an EroEI of less thanunity may still be economic to extract.

28 Studies based upon embodied energy have fallen out of favour since the 1980s, when they were often associated withsomewhat controversial ‘energy theories of value’ (Söllner, 1997). But there is no necessary link between these theories and useof ‘embodied energy’ estimates in empirical research. Consideration of rebound effects may provide a motivation for revivingthis area of research.

S4134 12/10/07 13:56 Page 43

15

20 25

30

Very similar effects will result from energyefficiency improvements by producers.For example, energy efficiencyimprovements in steel production shouldreduce the cost of steel and (assumingthese cost reductions are passed on inlower product prices)29 reduce the inputcosts of manufacturers that use steel. Thisin turn should reduce the cost of steelproducts and increase demand for thoseproducts. Such improvements could, forexample, lower the cost of passengercars, increase the demand for car traveland thereby increase demand for motor-fuel.

This example demonstrates how energyefficiency improvements could lead to aseries of adjustments in the prices andquantities of goods and services suppliedthroughout an economy. If the energyefficiency improvements are widespread,the price of energy intensive goods andservices may fall to a greater extent thanthat of non-energy intensive goods andservices, thereby encouraging consumerdemand to shift towards the former. Ifenergy demand is reduced, the resultingfall in energy prices will encourage greaterenergy consumption by producers andconsumers and will feed through intolower product prices, thereby encouragingfurther shifts towards energy intensivecommodities. Reductions in both energyprices and product prices will increaseconsumers’ real income, therebyincreasing demand for products,

encouraging investment, stimulatingeconomic growth and further stimulatingthe demand for energy. In somecircumstances, such improvements couldalso change trade patterns andinternational energy prices and thereforeimpact on energy consumption in othercountries.

A number of analysts have claimed thatthe secondary effects from energyefficiency improvements in consumertechnologies are relatively small (Lovins,et al., 1988; Greening and Greene, 1998;Schipper and Grubb, 2000). This isbecause energy makes up a small share oftotal consumer expenditure and theenergy content of most other goods andservices is also small. For example,suppose energy efficiency improvementsreduce natural gas consumption per unitof space heated by 10%. If there is nodirect rebound effect, consumers willreduce expenditure on natural gas forspace heating by 10%. If natural gas forheating accounts for 5% of totalconsumer expenditure, consumers willexperience a 0.5% increase in their realdisposable income. If all of this werespent on motor-fuel for additional cartravel, the net energy savings (in kWhthermal content) will depend upon theratio of natural gas prices to motor-fuelprices, and could in principle be more orless than one.30 In practice, however,motor-fuel only accounts for a portion ofthe total cost of car travel and car travel

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

44

15

20 25

30

29 Product prices will only fall if a sufficiently large number of domestic firms within the sector benefit from the energy efficiencyimprovement and will be limited by the extent to which the product market is exposed to international competition.

30 Chalkley et al. (2001) provide an example of the replacement of an inefficient (C-rated) refrigerator with an efficient (A-rated)model. Lifetime carbon savings for the refrigerator are estimated at 1645 kgCO2 and lifetime cost savings at £120.57. If thesecost savings were spent wholly on gasoline, the indirect CO2 emissions would be 358 kg, giving an indirect rebound effect, incarbon terms, of 22%. However, spending all of the cost savings on gasoline is unrealistic.

S4134 12/10/07 13:56 Page 44

15

20 25

30

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

45

only accounts for a portion of totalconsumer expenditure. For the greatmajority of goods and services, input-output data suggests that the effectiveexpenditure on energy should be less than15% of the total expenditure. Hence, bythis logic, the secondary effect should beonly around one tenth of the direct effect(Greening and Greene, 1998).

Entirely analogous arguments apply to thesecondary effects for producers, such asin the steel example above. Since energyforms a small share of total productioncosts for most firms and sectors (typically<3%) and since intermediate goods forma small share of the total costs of mostfinal products, the product of thesesuggests an indirect effect that is muchsmaller than the direct effect (Greeningand Greene, 1998).

However, while plausible, thesearguments are not supported by theresults of several of the quantitativestudies discussed below. Moreover, theyassume that the only effect of the energyefficiency improvement is to reduceexpenditure on energy. But as argued inSection 2, improvements in the energyefficiency of production processes arefrequently associated with improvementsin the productivity of capital and labour aswell and therefore lead to cost savingsthat exceed the savings in energy costsalone. In some cases, similar argumentsmay apply to energy efficiencyimprovements by consumers: forexample, a shift from car travel to cyclingcould save on depreciation andmaintenance costs for vehicles as well asmotor-fuel costs (Alfredsson, 2004). Inthese circumstances, the secondaryeffects that result from the adoption of a

particular technology could be substantialand may even exceed the direct energysavings.

4.3 Evidence for limits tosubstitutionSome indication of the importance ofembodied energy may be obtained fromestimates of the own-price elasticity ofaggregate primary, secondary or finalenergy demand. In principle, thismeasures the scope for substitutingcapital, labour and materials for energy,while holding output constant. Mostenergy price elasticities are estimated atthe level of individual sectors andtherefore do not reflect all the embodiedenergy associated with capital, labour andmaterials inputs. Since the own-priceelasticity of aggregate energy reflects thisindirect energy consumption, it should inprinciple be smaller than a weightedaverage of energy demand elasticitieswithin each sector. However, theaggregate elasticity may also reflectprice-induced changes in economicstructure and product mix which inprinciple could make it larger than theaverage of sectoral elasticities (Sweeney,1984). These two mechanisms thereforeact in opposition.

Based in part upon modelling studies,Sweeney (1984) puts the long-runelasticity of demand for primary energy inthe range -0.25 to -0.6. In contrast,Kaufmann (1992) uses econometricanalysis to propose a range from -0.05 to-0.39, while Hong (1983) estimates avalue of -0.05 for the US economy. A lowvalue for this elasticity may indicate a

S4134 12/10/07 13:56 Page 45

15

20 25

30

limited scope for substitution and hencethe potential for large indirect reboundeffects.31 But this interpretation is notstraightforward, since direct reboundeffects also contribute to the behaviourbeing measured. Also, measures of thequantity and price of ‘aggregate energy’are sensitive to the methods chosen foraggregating the prices and quantities ofindividual energy carriers, while the priceelasticity will also depend upon theparticular composition of price changes(e.g. increases in oil prices relative to gas)(EMF 4 Working Group, 1981). Inparticular, when different energy typesare weighted by their relative marginalproductivity, the estimated elasticitiestend to be lower (Hong, 1983) As a result,the available estimates of aggregate priceelasticities may be insufficiently precise toprovide much indication of the magnitudeof indirect rebound effects.

Relatively few empirical studies haveinvestigated the embodied energyassociated with specific energy efficiencyimprovements and those that have appearto focus disproportionately upon domesticbuildings. In a rare study of energyefficiency improvements by producers,Kaufmann and Azary Lee (1990) estimatethat, in the US forest products industryover the period 1954 to 1984, theembodied energy associated with capitalequipment offset the direct energysavings from that equipment by as much

as 83% (Box 4.1). But since theirmethodology is crude and the resultsspecific to the US context, this studyprovides little indication of the magnitudeof these effects more generally.

Estimates of the embodied energy ofdifferent categories of goods and servicescan be obtained from input-outputanalysis, life-cycle analysis (LCA) or acombination of the two (Chapman, 1974;Herendeen and Tanak, 1976; Kok, et al.,2006). A full life-cycle analysis is timeconsuming to conduct and must addressproblems of ‘truncation’ (i.e. uncertaintyover the appropriate system boundary)32

and joint production (i.e. how to attributeenergy consumption to two or moreproducts from a single sector) (Leach,1975; Lenzen and Dey, 2000). Hence,many studies combine standard economicinput-output tables with additionalinformation on the energy consumption ofindividual sectors, to give acomprehensive and reasonably accuraterepresentation of the direct and indirectenergy required to produce ratheraggregate categories of goods andservices. More detailed, LCA-basedestimates are available for individualproducts such as building materials, butresults vary widely from one context toanother depending upon factors such asthe fuel mix for primary energy supply(Sartori and Hestnes, 2007).Th

e R

ebou

nd E

ffect

: an

asse

ssm

ent

of t

he e

vide

nce

for

econ

omy-

wid

e en

ergy

sav

ings

from

impr

oved

ene

rgy

effic

ienc

y

46

15

20 25

30

31 This is in contrast to the own-price elasticity of energy demand for an individual energy service, where high values mayindicate the potential for large direct rebound effects.

32 For example, should the indirect energy costs of a building also include the energy used to make the structural steel and minethe iron ore used to make the girders? This is referred to as the truncation problem because there is no standard procedure fordetermining when energy costs become small enough to neglect

S4134 12/10/07 13:56 Page 46

15

20 25

30

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

47

Kaufmann and Azary Lee (1990) examined the embodied energy associated with energyefficiency improvements in the US forest products industry over the period 1954 to1984. First, they estimated a production function for the output of this industry and usedthis to derive the ‘marginal rate of technical substitution’ (MRTS) between capital andenergy in a given year - in other words, the amount of gross fixed capital that was usedto substitute for a thermal unit of energy in that year. Second, they approximated theembodied energy associated with that capital by means of the aggregate energy/GDPratio for the US economy in that year - hence ignoring the particular type of capital used,as well as the difference between the energy intensity of the capital producing sectorsand that of the economy as a whole. The product of these two variables gave anestimate of the indirect energy consumption associated with the gross capital stock usedto substitute for a unit of energy. This was then multiplied by a depreciation rate to givethe energy associated with the capital services used to substitute for a unit of energy.

Finally, they compared the estimated indirect energy consumption with the direct energysavings in the forest products sector in each year. Their results showed that the indirectenergy consumption of capital offset the direct savings by between 18 and 83% overthe period in question, with the net energy savings generally decreasing over time. Theprimary source of the variation was the increase in the MRTS over time, implying thatan increasing amount of capital was being used to substitute for a unit of energy.However, the results were also influenced by the high energy/GDP ratio of the USeconomy, which is approximately twice that of many European countries. Overall, thecalculations suggest that the substitution reduced aggregate US energy consumption,but by much less than a sector-based analysis would suggest. Also, their approach didnot take into account any secondary effects resulting from the energy efficiencyimprovements.

The simplicity of this approach suggests the scope for further development and widerapplication. Accuracy could be considerably improved by the use of more flexibleproduction functions and more precise estimates for the indirect energy consumptionassociated with specific types and vintages of capital goods. However, to date no otherauthors appear to have applied this approach to particular industrial sectors or to haverelated it to the broader debate on the rebound effect.

Box 4.1 Limits to substitution for producers

S4134 12/10/07 13:56 Page 47

15

20 25

30

As an illustration, Sartori and Hestnes

(2007) reviewed 60 case studies of

buildings, and found that the share of

embodied energy in life-cycle energy

consumption ranged between 9 and 46%

for low energy buildings and between 2

and 38% for conventional buildings – with

the wide range reflecting different building

types, material choices and climatic

conditions. Two studies that controlled for

these variables found that low energy

designs could achieve substantial

reductions in operating energy

consumption with relatively small

increases in embodied energy, leading to

‘payback periods’ for energy saving of as

little as one year (Feist, 1996; Winther

and Hestnes, 1999). Similar calculations

were performed by the Royal Commission

on Environmental Pollution (2007), who

estimate a 15 year simple payback (in

energy terms) for low energy new build

houses in the UK. However, Casals (2006)

shows how the embodied energy of such

buildings could offset operational energy

savings, even with an assumed 100 year

lifetime. Such calculations typically

neglect differences in energy quality and

the results are sensitive to context, design

and building type. However, the increasing

availability of embodied energy

coefficients at a national level (e.g. Alcorn

and Baird (1996)) suggests the scope for

greater use of such estimates in policy

evaluation.

In the case of existing buildings, several

studies suggest that retrofits of thermal

insulation pay for themselves in terms of

energy savings within a few months

(compared to a useful life in excess of 25

years), while the corresponding period for

double glazing is several years. In other

cases, for example condensing boilers

compared to conventional boilers, the

variation of embodied energy within

individual categories of boiler exceeds the

difference between them. Hence, the

contribution of embodied energy to the

economy-wide rebound effect appears to

vary widely from one situation to another

and is inversely proportional to the

lifetime of the energy saving measure.

But the patchy nature of this evidence

base, the lack of systematic comparisons

of energy efficiency options and the

dependence of the results on particular

contexts all make it difficult to draw any

general conclusions.

4.4 Evidence for secondaryeffectsBy combining estimates of the embodiedenergy associated with differentcategories of goods and services withsurvey data on household consumptionpatterns, it is possible to estimate thetotal (direct plus indirect) energyconsumption of different types ofhousehold; together with the indirectenergy consumption associated withparticular categories of expenditure (Kok,et al., 2006). It is often found that theindirect energy consumption ofhouseholds exceeds the directconsumption. Moreover, while indirectenergy consumption increases withincome, direct energy consumption showssigns of saturation - suggesting that

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

48

15

20 25

30

S4134 12/10/07 13:56 Page 48

15

20 25

30

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

49

indirect energy consumption is becomingincreasingly important over time.33

If this data is available at a sufficientlydisaggregated level, it could also be usedto estimate the secondary effectsassociated with energy efficiencyimprovements by households - providedthat additional information is available oneither the cross price elasticity betweendifferent product and service categories,or the marginal propensity to spend34 ofdifferent income groups. By combiningthe estimates of embodied energy andsecondary effects, an estimate of the totalindirect rebound effect may be obtained.Such approaches are ‘static in that theydo not capture the full range of price andquantity adjustments, but couldnevertheless be informative.35 However,of the 19 studies in this area reviewed byKok, et al. (2006), only three wereconsidered to have sufficient detail toallow the investigation of such micro-levelchanges – largely because they combinedinput-output with LCA data (Bullard, et al.,1978). Hence, estimation of secondaryeffects by this route appears to be in itsinfancy.

Three studies that use this generalapproach are summarised here. First,Brännlund et al. (2007) examine the

effect of a 20% improvement in theenergy efficiency of personal transport (allmodes) and space heating in Sweden.They estimate an econometric model ofaggregate household expenditure, inwhich the share of total expenditure forthirteen types of good or service isexpressed as a function of the totalbudget, the price of each good or serviceand an overall price index. This allows theown-price, cross-price and incomeelasticities of each good or service to beestimated.36 Energy efficiencyimprovements reduce the cost oftransport and heating and lead tosubstitution and income effects thatchange overall demand patterns (e.g.improvements in transport efficiency areestimated to increase demand for clothesbut to decrease demand for beverages).By combining these estimated changes indemand patterns with CO2 emissioncoefficients for each category of good andservice (based upon estimates of directand indirect energy consumption)Brännlund et al. find that energyefficiency improvements in transport andheating lead to (direct + indirect)rebound effects (in carbon terms) of120% and 170% respectively.

Brännlund et al.’s results are heavilydependent on the assumed carbon

33 Results vary widely with country, time period and methodology. For example, Herendeen (1978) found that indirect energyconsumption in Norway accounted for one third of total energy consumption for a poor family and approximately two thirds fora rich family. Vringer and Blok (1995) found that 54% of total energy demand in Dutch households was indirect, while Lenzen(1998) found that 30% of total energy demand in Australian households was indirect.

34 Defined as the change in expenditure on a particular product or service, divided by the change in total expenditure. Themarginal propensity to spend on different goods and services varies with income and it is an empirical question as to whetherthe associated indirect energy consumption is larger or smaller at higher levels of income. However, the greater use of energyintensive travel options by high income groups (notably flying) could be significant in some cases.

35 In technical terms, these provide a partial equilibrium analysis, as distinct from the general equilibrium analysis provided byCGE models.

36 Brännlund et al. employ Almost Ideal Demand (AID) model, which has been shown to have a number of advantages overother models of consumer demand (Deaton and Mulbauer, 1980; Xiao, et al., 2007). The model relies on the assumption of‘staged-budgeting’: for example consumers are assumed to first decide on the proportion of their budget to spend on transport,and then to decide how to allocate their transport budget between different modes. While analytically convenient, thisassumption is likely to be flawed.

S4134 12/10/07 13:56 Page 49

15

20 25

30

emission coefficients, but the source ofthese is not made explicit. The results alsocontradict the econometric evidence ondirect rebound effects, since carbonemissions for heating and transport areestimated to increase. Furthermore,Brännlund et al. use an iterativeestimation procedure, but only presentthe results from the first estimation step.This weakness is overcome by Mizobuchi(2007), who follows a very similarapproach to Brännlund et al, but appliedto Japanese households. Despite thedifferences in data sources and estimationprocedures, the estimated rebound effectsare broadly the same. However, Mizobuchialso examines the effect of the additionalcapital cost of energy efficient equipmentand finds that these reduce the reboundeffect significantly.

The third example adopts a differentapproach, using data on the marginalpropensity to spend of different incomegroups in Sweden. Alfredsson (2004)calculates the direct and indirect energyconsequences of ‘greener’ consumptionpatterns, which include both technicalchanges, such as buying a more fuel-efficient car, and behavioural changessuch as car sharing. In the case of‘greener’ food consumption (e.g. shiftstowards a vegetarian diet), the totalenergy consumption associated with fooditems is reduced by around 5% and totalexpenditure on food items is reduced by15%. But the re-spending of this moneyon a variety of items, notably travel andrecreation, leads to indirect energyconsumption that more than offsets theoriginal energy savings (i.e. backfire). Theresults for a shift towards ‘greener’ travelpatterns are less dramatic, but the

secondary effects from re-spendingreduce the overall energy savings byalmost one third. A comprehensive switchto green consumption patterns in travel,food and housing is estimated to have arebound effect of 35%.

Secondary effects are relatively large inthis example because ‘green’ consumptionreduces expenditure on more than energyalone. Also, the results from such studiesdepend upon the methodology andassumptions used, as well as the types ofhousehold analysed and the particularshifts in consumption patterns that areexplored. For example, a more recentstudy (Kanyama, et al., 2006) using asimilar model and approach to Alfredsson,but employing Swedish rather than Dutchdata on energy intensity, finds that a shiftto ‘green’ food consumption could reduceoverall energy consumption. Closerexamination reveals that this resultfollows largely from the assumption thatgreener diets are more expensive (owingto the higher cost of locally producedorganic food), thereby leading to anegative ‘re-spending’ effect.

In sum, the potential of embodied energyapproaches to estimating secondaryeffects has yet to be fully explored. Whilethe studies reviewed here suggest thatsecondary effects may sometimes belarger than commonly assumed, theconclusions may change oncemethodological weaknesses areaddressed or a different choice ofindependent variable is made. Hence, atpresent the available evidence is too smallto permit any general conclusions to bedrawn.

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

50

15

20 25

30

S4134 12/10/07 13:56 Page 50

15

20 25

30

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

51

4.5 Evidence for economy-wide effects from generalequilibrium modelsThe economy-wide rebound effectrepresents the sum of the direct andindirect rebound effects and will dependupon the size, nature and location of theenergy efficiency improvements. Giventhe number of confounding variables, it islikely to be very difficult to estimatethrough the econometric analysis ofsecondary data.

One approach, although it would notpermit quantification of the overall effect,would be to identify those variables thatshould influence the size of direct andindirect effects and to estimate theseempirically. Direct rebound effects forproducers, for example, should beinfluenced by the own-price elasticity ofdemand for the relevant products, theshare of energy in the total cost ofproduction and the extent to which thecheaper energy services are able tosubstitute for capital, labour and materials(Saunders, 1992; Allan, et al., 2006).Rebound effects may be expected to belarger in energy intensive sectors and alsowhere the input mix is fairly flexible andwhere the demand for products isrelatively price-elastic.

An alternative approach is to estimate themagnitude of economy-wide effectsthrough energy-economic models of themacro-economy (Grepperud andRasmussen, 2004). Despite thewidespread use of such models within

energy studies (Bhattacharyya, 1996), itis only recently that attempts have beenmade to quantify economy-wide reboundeffects in this way.37 The literature istherefore extremely sparse, but nowincludes two insightful studiescommissioned by the UK government(Allan, et al., 2006; Barker and Foxon,2006). A key distinction is betweenComputable General Equilibrium (CGE)models of the macroeconomy and thosebased upon econometrics. This sectiondiscusses some results from CGE models,while Section 4.5 discusses an applicationof a macro-econometric model to therebound effect.

Computable General Equilibrium (CGE)models are widely used in theinvestigation of energy and climate policy,partly as a consequence of the readyavailability of modelling frameworks andthe associated benchmark data. Thisapproach is informed by neoclassicaleconomic theory, but can deal withcircumstances that are too complex foranalytical solutions.

CGE models are calibrated to reflect thestructural and behavioural characteristicsof particular economies and in principlecan indicate the approximate order ofmagnitude of direct and indirect reboundeffects from specific energy efficiencyimprovements. A CGE model should allowthe impacts of such improvements to beisolated, since the counterfactual is simplya model run without any changes inenergy efficiency, as well as allowing therebound effect to be decomposed into itsconstituent components, such as

37 A special edition of the Energy Policy journal published in 2000 provides a invaluable account of the state of the debate atthat time, but does not include any modelling estimates of economy-wide rebound effects (Schipper, 2000)

S4134 12/10/07 13:56 Page 51

15

20 25

30

substitution and output effects. Inprinciple, CGE models also provide scopefor sensitivity analysis, although inpractice this appears to be rare.

CGE models have a number of limitations(see Box 4.2) that have led many authorsto question their realism and policyrelevance (Barker, 2005). Whiledevelopments in CGE methodology arebeginning to overcome some of theseweaknesses, most remain. Also, thepredictive power of such models is rarelytested and different models appear toproduce widely varying results for similarpolicy questions (Conrad, 1999). Hence,while CGE models may provide valuableinsights, the quantitative results of suchmodels should be interpreted with caution.

Eight CGE modelling studies of reboundeffects have been identified and reviewed(Table 4.1). These models varyconsiderably in terms of the productionfunctions used, the manner in whichdifferent inputs are combined (the‘nesting’ structure), the assumed scopefor substitution between different inputs,the treatment of labour supply, themanner in which government savings arerecycled and other key parameters. Thesestudies also vary widely in their simulationof energy efficiency improvements, withsome introducing an across the boardimprovement and others introducing aspecific improvement in an individualsector, or combination of sectors. Thisdiversity, combined with the limitednumber of studies available again makes itdifficult to draw any general conclusions.

The most notable result is that all of thestudies find economy-wide reboundeffects to be greater than 37% and moststudies show either large rebounds

(>50%) or backfire. The latter was foundin two studies of economies in whichenergy forms an important export andimport commodity, suggesting that this isa potentially important and hithertoneglected variable. Allan et al. (2006) finda long-term rebound effect of 31% fromacross-the-board improvements in theenergy efficiency of UK production sectors,including primary energy supply. Thisstudy is summarised in Box 4.3.

The results suggest that the magnitude ofrebound effect depends upon a wide rangeof variables. While Saunders (1992) andothers stress the importance of the‘elasticity of substitution’ between energyand other inputs in production (Box 4.4),other characteristics such as the elasticityof supply of capital and labour, the own-price elasticity of demand for the productof each sector, the energy intensity ofproducing sectors, the scope forsubstitution between differentconsumption goods, the income elasticityof the demand for goods and the mannerin which government revenue isredistributed are also potentiallyimportant. The contrasting results ofHanley et al. (2005) for the Scottisheconomy (rebound >100%) and Allan etal. (2006) for the UK economy (rebound ~37%) are revealing, since the assumedimprovement in energy efficiency is thesame in both cases (5% in all productionsectors). The difference is primarily due tothe relative sensitivity of export demandto changes in energy efficiency. While bothstudies assume that energy efficiencyimprovements are made in electricitygeneration, it is only in the Scottish casethat this leads to a substantial increase inelectricity exports – and hence in domesticenergy consumption.

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

52

15

20 25

30

S4134 12/10/07 13:56 Page 52

15

20 25

30

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

53

• Market and behavioural assumptions: CGE models are based upon a number of standard neo-classical assumptions (e.g. utility maximization; perfect competition; constant returns to scalein production; etc.) that are poorly supported by empirical evidence. In particular, thepossibility of ‘win-win’ policies, such as those aimed at encouraging energy efficiency, may beexcluded if an economy is assumed to be at an optimal equilibrium.

• Functional Forms: The choice of utility and production functions is governed by the need forconvenience and solvability, rather than realism and generally imposes restrictions on thebehaviour of the economy being modelled. The choice of a different production function canlead models to predict different outcomes to an identical policy change. The most commonapproach is the ‘nested constant elasticity of substitution’ (CES) production function, but themanner in which this is implemented varies widely from one model to another. This structurealso assumes that the optimum ratio of two inputs that are grouped together in a ‘nest’ isunaffected by either the level or price of other inputs (‘separability’). However, this assumptionis generally not supported by empirical evidence (Frondel and Schmidt, 2004).

• Calibration and parameters: CGE models are calibrated to benchmark data from a single yearand adjustments are made to the data to ensure that the equilibrium assumption holds. Thechoice of base year for calibration is somewhat arbitrary and may influence the results. Someparameter values will be determined through this calibration, while others will be set externallyon the basis of econometric literature. However, the time periods/regions/sectors on which thisliterature is based may not be appropriate to the model application, and the process ofcompiling parameter values may not always be transparent. Assumed parameter values couldhave a significant influence on the model results, but the calibration approach provides noinformation on the statistical reliability of individual estimates. Sensitivity tests are possible,but these are frequently confined to a small number of relevant parameters. Also, the modelingtypically rests on the implicit assumption that such parameters will remain stable over time.

• Sensitivity and transparency: There exists considerable variation between CGE models so thatcare needs to taken when comparing results. Changing some assumptions can sometimesgenerate very different simulated outcomes. The model results can be driven by assumptionsthat are not apparent to a reader not acquainted with the model. But CGE models are notnecessarily a ‘black box’: transparency may be considerably improved by providing informationon key features and assumptions and explaining model results with reference to economictheory.

• Treatment of Technical Change: In most CGE models, technological change is assumed to becostless and exogenous and thus not related to or determined from within the system. Acommon approach is to assume that energy efficiency improves at a constant rate over time,but this fails to capture price/policy induced innovation and other characteristics of endogenoustechnical change.

• Time scales: Many (but not all) CGE models simulate equilibrium states of the economy anddo not consider the dynamics of adjustment and the associated costs. The solutions areassumed to refer to the ‘long-run’, but the time period that this represents may be unclear. Forexample, transport and building infrastructures take much longer to adjust than other types ofcapital equipment.

Box 4.2 Limitations of CGE models

S4134 12/10/07 13:56 Page 53

15

20 25

30Th

e R

ebou

nd E

ffect

: an

asse

ssm

ent

of t

he e

vide

nce

for

econ

omy-

wid

e en

ergy

sav

ings

from

impr

oved

ene

rgy

effic

ienc

y

54

15

20 25

30

Table 4.1 Summary of CGE studies of rebound effects

Author/

Date

Country

or

region

Nesting

structure

ESUB

with

energy

Assumed

energy

efficiency

improvements

Estimated

rebound

effect

Comments

Sem

boja

,

1994a

Ken

yaEle

c,

fuel

,K,

L

1.0

Tw

osc

enari

os:

an

impro

vem

ent

of

energ

ypro

duct

ion

effici

ency

and

an

impro

vem

ent

inen

erg

yuse

effici

ency

.

>100%

Intu

itiv

epre

senta

tion,

no

sensi

tivi

tyte

sts,

lack

of

transp

are

ncy

.

Dufo

urn

aud

etal,

1994

Sudan

Utilit

y

funct

ions

0.2

and

0.4

.

100%

,150%

and

200%

impro

vem

ent

ineff

icie

ncy

of

wood-

burn

ing

stoves

47-77%.

Models

effici

ency

impro

vem

ents

indom

est

icst

oves

.W

ide

range

of

sensi

tivity

test

sand

good

exp

lanation

of

the

fact

ors

at

work

.

Vik

stro

m,

2004

Sw

eden

(KE)L

Valu

es

range

from

0.0

7

to0.8

7.

15%

incr

ease

inen

erg

yeff

icie

ncy

in

non-e

nerg

yse

ctors

and

12%

incr

ease

inen

erg

yse

ctors

.

50-60%

Applie

sto

1957-1

962

peri

od

inw

hic

hknow

nch

anges

inener

gy

effici

ency

pro

duct

ivity

and

stru

cture

are

com

bin

edin

turn

.R

esu

lts

apply

only

toen

erg

yeff

icie

ncy

com

ponen

t

Wash

ida,

2004

Japan

(KL)

E0.5

.1%

inall

sect

ors

model

led

as

change

ineff

icie

ncy

fact

or

for

use

of

energ

yin

pro

duct

ion

53%in

central

scenario

Pre

senta

tion

uncl

ear,

although

there

isso

me

sensi

tivity

analy

sis,

incl

udin

gva

ryin

g

elast

icity

of

subst

itution

from

0.3

to0.7

join

tly

with

oth

er

para

mete

rs.

Reb

ound

eff

ect

incr

ease

sas

energ

y/va

lue

added

,la

bour/

capital

and

level

of

energ

yco

mposi

te

subst

itution

ela

stic

itie

sin

crea

se.

Gre

pper

ud

and

Rasm

uss

en

2004

Norw

ay

(KE)L

Betw

een

0&

1

Doublin

gof

gro

wth

rate

sof

energ

y

pro

duct

ivity.

Four

sect

ors

have

elect

rici

tyeff

icie

ncy

double

d,

&tw

o

have

oil

eff

icie

ncy

double

d.

Smallfor

oilbut

>100%for

electricity

Model

issi

mula

ted

dynam

ically

with

aco

unte

rfact

ual

case

inw

hic

hpro

ject

ions

of

world

eco

nom

icgro

wth

,la

bour

forc

egro

wth

,te

chnolo

gic

al

pro

gre

ssand

net

fore

ign

deb

tare

ass

um

eduntil2050.

Glo

msr

ød

and

Taoyu

an,

2005

Chin

aEle

c,

fuel

,K,

L

1.0

Busi

ness

-as-

usu

al

scen

ari

o

com

pare

dto

case

wher

eco

stle

ss

invest

men

tsgener

ate

incr

ease

d

invest

men

tsand

pro

duct

ivit

yin

coal

clea

nin

g-

low

ering

pri

ceand

incr

easi

ng

supply

>100%

Coalin

tensi

ve

sect

ors

benef

it,

as

does

whole

econom

ydue

tohig

huse

of

coalin

pri

mary

energ

yco

nsu

mption.

Als

oex

am

ines

case

sw

her

eco

aluse

issu

bje

ctto

emis

sions

tax.

Hanle

yet

al,

2005

Sco

tlan

d

(KL)

(EM

)0.3

.5%

impro

vem

ent

ineff

icie

ncy

of

energ

yuse

acr

oss

all

pro

duct

ion

sect

ors

>100%

Reg

ion

issi

gnific

ant

elect

rici

tyexp

ort

er

and

resu

ltdep

ends

inpart

on

incr

ease

d

elect

rici

tyexp

ort

s

Alla

netal,

2006

UK

(KL)

(EM

)0.3

5%

impro

vem

ent

ineff

icie

ncy

of

energ

yuse

acr

oss

all

pro

duct

ion

sect

ors

(incl

udin

gener

gy

sect

ors

).

37%in

central

scenario

See

Box

4.3

Note

s: T

he

pro

duct

ion f

unct

ions

com

bin

e in

puts

into

pai

rs,

or

‘nes

ts’.

For

exam

ple

, a

nes

ted p

roduct

ion f

unct

ion w

ith c

apital

(K),

lab

our

(L)

and e

ner

gy

(E)

inputs

, co

uld

take

one

of th

ree

form

s, n

amel

y: K

(LE);

(KL)

E;

(KE)L

. ESU

B m

eans

the

elas

tici

ty o

f su

bst

itution b

etw

een e

ner

gy

and o

ther

inputs

. In

terp

reta

tion d

epen

ds

upon t

he

nes

ting

stru

cture

. Fo

r ex

ample

, fo

r K(L

E)

ESU

B r

efer

s to

the

elas

tici

ty o

f su

bst

itution b

etw

een L

and E

, w

hile

for

(KL)

E it

refe

rs t

o e

last

icity

of

subst

itution b

etw

een (

KL)

and E

.

S4134 12/10/07 13:56 Page 54

15

20 25

30

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

55

Allan et al. (2006) estimate economy-wide rebound effects for the UK following a 5%across-the-board improvement in the efficiency of energy use in all production sectors.Since their model allows for the gradual updating of capital stocks, they are able toestimate a short run rebound effect of 50% and a long-run effect of 31% (whichcontradicts most of the literature, which estimates long-run effects to be larger).

The energy efficiency improvements increase long-run GDP by 0.17% and employmentby 0.21%. They have a proportionally greater impact on the competitiveness of energyintensive sectors which is passed through in lower product prices despite a 0.3%increase in real wages. Output is increased in all sectors, with the iron and steel andpulp and paper sectors benefiting the most with long-run increases of 0.67% and 0.46%respectively. In contrast, the output of the oil refining and electricity industries (i.e. oiland electricity demand) is reduced, with the price of conventional electricity falling by24% in the long-run. This fall in energy prices contributes a significant proportion of theoverall rebound effect and results from both cost reductions in energy production - dueto the energy efficiency improvements - and reduced energy demand.

Allan et al. are using an economic rather than thermodynamic measure of energyefficiency, but a 5% improvement may nevertheless be impractical for industries suchas electricity generation and oil refining which are operating close to thermodynamiclimits. It would also require major new investment and take time to be achieved. Theresults suggest that the economy-wide rebound effect would be smaller if energyefficiency improvements were confined to energy users, but this scenario was notinvestigated. Moreover, the results suggest that energy efficiency improvements in theenergy supply industry may be associated with large rebound effects.

A notable feature of this study is the use of sensitivity tests. Varying the assumedelasticity of substitution between energy and non-energy inputs between 0.1 and 0.7(compared to a baseline value of 0.3) had only a small impact on economic output (from0.16% in the low elasticity case to 0.10% in the high case), but a major impact on therebound effect. This varied from 7% in the low case to 60% in the high case. Grepperudand Rasmussen (2004) report similar results, which highlights the importance of thisparameter for CGE simulations. Unfortunately, as shown in Box 4.4, the empirical basisfor this parameter is weak.

Varying the elasticity of demand for exports was found to have only a small impact onGDP and energy demand, suggesting that the energy efficiency improvements had onlya small impact on the international competitiveness of the relevant industries. However,different treatments of the additional tax revenue were found to be important, withrecycling through lower taxes increasing the rebound effect from 31% to 40%.

Box 4.3 CGE estimates of economy-wide rebound effects for the UK

S4134 12/10/07 13:56 Page 55

15

20 25

30Th

e R

ebou

nd E

ffect

: an

asse

ssm

ent

of t

he e

vide

nce

for

econ

omy-

wid

e en

ergy

sav

ings

from

impr

oved

ene

rgy

effic

ienc

y

56

15

20 25

30

All but one of the models explore theimplications of energy efficiencyimprovements in production sectors andthe CGE literature offers relatively littleinsight into the implications of energyefficiency improvements in consumergoods. Since there are differences acrossincome groups, this would require a muchgreater detail on the demand side of theCGE models than is commonly the case,together with more accuraterepresentations of household behaviour.

CGE models also simulate energyefficiency improvements as pure ‘energyaugmenting technical change’, which isassumed to be costless. Of the studiesreviewed here, only Allan et al. (2006)consider the implications of additionalcosts associated with energy efficiencyimprovements and they find that reboundeffects are correspondingly reduced.

In summary, given the small number ofstudies available, the diversity ofapproaches and the methodological

A key parameter in CGE models is the so-called elasticity of substitution betweendifferent inputs. This determines the extent to which one input can substitute foranother while keeping output constant. The elasticity of substitution between energyand other inputs has a strong influence on the estimated magnitude of rebound effects(Grepperud and Rasmussen, 2004).

There are a large number of empirical estimates of elasticities of substitution betweendifferent inputs, within different countries and sectors and over different periods of time.In principle, this literature could be used as a basis for selecting appropriate parametervalues in individual CGE models. In practice, however, there is only a tenuousrelationship between the parameters estimated by empirical studies and those assumedby the models. In particular, CGE models:

• use different types of production function from those estimated within empiricalstudies;

• use different definitions of the elasticity of substitution from those estimated withempirical studies;

• combine inputs into nests while most empirical studies do not;

• assume that the scope for substitution within a nest is independent of the level orprices of other inputs, while most empirical studies do not;

• make assumptions about the elasticity of substitution between different nests, whilemost empirical studies provide estimates of the elasticity of substitution betweenindividual pairs of inputs.

In addition: the process of compiling parameter values is rarely transparent; sensitivitytests are uncommon; the empirical studies frequently apply to different sectors, timeperiods and levels of aggregation to those represented by the model; and differentmodels use widely different assumptions. These observations suggest that the empiricalbasis for most CGE models is relatively weak.

Box 4.4 Parameterising CGE models

S4134 12/10/07 13:56 Page 56

15

20 25

30

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

57

weaknesses of the CGE approach, it is notpossible to draw any general conclusionsregarding the size of economy-widerebound effect. Indeed, since the mostimportant insight from this literature isthat the economy-wide rebound effectvaries widely from one circumstance toanother, a general statement on the sizeof such effects may be misleading.Perhaps the most useful application of thisapproach would be to investigate thedeterminants of rebound effects moresystematically. It is worth noting,however, that the available studiessuggest that economy-wide effects arefrequently large and that the potential forbackfire cannot be ruled out. Moreover,these effects derive from ‘pure’ energyefficiency improvements and therefore donot rely upon simultaneous improvementsin the productivity of capital and labourinputs.

4.6 Evidence for economy-wide effects from macro-econometric modelsMacro-econometric models can overcomeseveral of the weaknesses of CGEmodelling while at the same time providinga greater level of disaggregation thatpermits the investigation of specificgovernment policies. Their usefulness canbe further enhanced if they can beeffectively linked to ‘bottom-up models’38

of particular sectors, such as electricitygeneration, where representation ofspecific technologies is desirable. Incontrast to their CGE counterparts, macro-econometric models do not rely uponrestrictive assumptions such as constantreturns to scale and perfect competition.They also replace the somewhat ad-hocuse of parameter estimates witheconometric equations estimated forindividual sectors which implicitly reflectnon-optimising behaviour, such as theapparent neglect of cost-effectiveopportunities to improve energy efficiency.However, this greater realism is achievedat the expense of greater complexity, moreonerous data requirements and highercosts in developing and maintaining suchmodels. This section describes the use ofone such model (MDM-E3)39 to exploreeconomy-wide rebound effects in the UK(Barker and Foxon, 2006; Barker, et al.,2007). At present, this appears to be theonly application of such models to theeconomy-wide rebound effect.

MDM-E3 combines time-series econometricrelationships and cross-sectional input-output relationships with detailedmodelling of the energy sector (Junankar,et al., 2007). It distinguishes 27investment sectors, 51 categories ofhousehold expenditure and 41industries/commodities. The modelincludes equations for consumption,investment, employment and trade foreach of the 41 industries, together withequations for wage rates, output and

38 ‘Bottom-up’ or engineering-economic models usually focus upon individual sectors and energy services and contain detailedinformation on the performance and cost of energy-using equipment. They simulate the ageing and replacement of thisequipment and seek to minimise the net cost of energy services given appropriate assumptions about individual andorganisational investment behaviour.

39 The Cambridge Multisectoral Dynamic Model of the UK energy-environment-economy (MDM-E3) is developed and maintainedby Cambridge Econometrics Ltd and also used by the Cambridge Centre for Climate Change Mitigation Research (4CMR) at theDepartment of Land Economy, University of Cambridge.

S4134 12/10/07 13:56 Page 57

15

20 25

30

commodity prices. The projected activitylevels are used as inputs to an energy sub-model, along with data on temperature andenergy prices and assumptions abouttechnical change. This leads to estimates oftotal energy demand, fuel and electricityuse by 13 categories of user and electricityprices. The sub-model then feeds thisinformation back into the main model byadjusting the input-output coefficients forproducers and consumer expenditures onenergy.40 The two models are runiteratively to reach a stable solution overthe projection period. Both the main modeland electricity sub-model use modern‘cointegration’ techniques to distinguishshort-term dynamic responses from long-term relationships (Engle and Granger,1987; Cambridge Econometrics, 2005)

In a study conducted on behalf of the UKgovernment, MDM-E3 was used to simulatethe macroeconomic impact of a number ofUK energy efficiency policies over theperiod 2000 to 2010 (Barker and Foxon,2006). The first stage was to developexogenous estimates of the ‘gross’ energysavings from each of these policies, the‘net’ energy savings which allow for theestimated direct rebound effects associatedwith each policy and the associatedinvestment costs. While Barker and Foxon(2006) state that the direct rebound effectfrom energy efficiency improvements byproducers was assumed to be zero, theoutput effects from such improvements aresubsequently modelled. Hence, it is thesubstitution effect from suchimprovements that is assumed to be zero.

The next stage was to run the model for abase case that included the full set of

energy efficiency policies, together withreference scenario that did not. Theestimated net energy savings wereincluded directly in the energy demandequations for each sector, the results ofwhich were then fed back into themacroeconomic model. Given theprojected lower energy demand, lowerenergy prices and hence lower energycosts for each sector, the model projecteda range of secondary effects includingincreased output and changes in tradepatterns. The estimated investment costsof the energy efficiency measures werealso taken into account by including themin the relevant investment equations.Overall, the cost-effective energy efficiencymeasures contributed to a 1.26% increasein GDP by 2010 (relative to the referencescenario) a 0.84% increase in employmentand a slight increase in imports. Energyconsumption was 13.8% lower than in thereference scenario, even after allowing foreconomy-wide rebound effects.

Indirect rebound effects were estimated asthe difference between the energy savingsprojected by the model and the estimatednet energy savings, expressed as apercentage of the estimated gross energysavings. The base run led to an estimatedindirect rebound effect of 11% by 2010.Closer inspection shows that indirectrebound effects were higher in the energyintensive industries (25%) and lower forhouseholds and transport (7%) (theopposite to what was assumed for directeffects). The primary source of the indirecteffects was substitution between energyand other goods by households, togetherwith increases in output by (particularlyenergy intensive) industry, which in turn

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

58

15

20 25

30

40 This contrasts with other model frameworks, in which projections from a macroeconomic model are simply used as inputs intoan energy sub-model, without any feedback from the latter.

S4134 12/10/07 13:56 Page 58

15

20 25

30

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

59

led to increased demand for both energyand energy-intensive intermediate goods.Increases in consumers’ real incomecontributed relatively small rebound effects(0.2%).

The direct rebound effects were estimatedto reduce overall energy savings by 15%,leading to an estimated economy-widerebound effect of 26% in 2010. Indirecteffects appeared relatively insensitive tovariations in energy and carbon prices. Theauthors claim that the results supportGrubb’s (1990) argument that policy-induced energy efficiency improvementsare associated with small rebound effectsbecause they focus on energy services witha low own-price elasticity as a consequenceof various market failures. This ismisleading, however: first, because thisargument applies to direct rebound effectsrather than the indirect effects beingmodelled; and second because theseservices should become more price elasticif the market failures are overcome.

The macro-econometric approachexemplified by MDM-E3 appears to offer amore robust approach to estimatingeconomy-wide rebound effects than CGEmodelling. Nevertheless, there are anumber of reasons why this particularstudy may underestimate economy-widerebound effects. First, while output effectsare estimated, the substitution between(cheaper) energy services and other inputswithin production sectors is ignored. Also,the direct rebound effects for consumershave to be estimated rather than modelled.Second, the modelling implicitly assumes‘pure’ energy efficiency improvements,with no associated improvements in theproductivity of other inputs. But if energyefficient technologies are commonlyassociated with such improvements,

rebound effects could be larger. Third, themodel does not reflect the indirect energyconsumption embodied within the energyefficient technologies themselves. Whilethe extra investment is taken into account,it appears to reduce the overall energyconsumption associated with investmentrather than increase it. Finally, the modelconfines attention to national energy useand ignores the indirect energyconsumption associated with increasedimports and tourism. This omission couldbe significant from a climate changeperspective, since it corresponds to ~40%of the extra domestic output.

In sum, while greater confidence can beplaced in this estimate of economy-widerebound effects than those from CGEmodels, there are a number of reasons whythis study may underestimate the fulleffect. Also, any similarity between thisresult and that of Allan et al. (2006) isspurious, since they use differentapproaches to model different types ofrebound effect from different types andsize of energy efficiency improvement indifferent sectors.

4.7 Summary • There are very few quantitative

estimates of indirect and economy-wide rebound effects and those thatare available have a number of flaws.While a number of methodologicalapproaches are available to estimatethese effects, the limited number ofstudies to date provides an insufficientbasis to draw any general conclusions.

• Techniques are available to estimatethe embodied energy associated withenergy efficiency improvements, but

S4134 12/10/07 13:56 Page 59

15

20 25

30

the potential of these techniques hasyet to be fully exploited. Most studiesfocus on domestic buildings anddemonstrate that embodied energyassociated with energy efficientbuildings varies widely betweendifferent building types, designs andcontexts. In most cases, the embodiedenergy is more than outweighed by thedirect energy savings during thebuilding life-cycle, but the contributionof embodied energy to the economy-wide rebound effect is generallyignored.

• The secondary effects associated withenergy efficiency improvements byhouseholds may be estimated througha combination of embodied energyanalysis and econometric models ofconsumer behaviour. Only a handful ofstudies have adopted this approach,and these give estimates of between33% and 170% for direct andsecondary effects combined. However,these studies have a number ofweaknesses, and the results are verydependent upon the particularapplication as well as the data andmethodologies employed.

• CGE studies suggest that themagnitude of economy-wide reboundeffects depend very much upon thesector where the energy efficiencyimprovements take place and aresensitive to a number of variables.While only a handful of studies areavailable, they suggest that economy-wide rebound effects may frequentlyexceed 50% and the potential forbackfire cannot be ruled out. Moreover,these rebound effects derive from‘pure’ energy efficiency improvementsand therefore do not rely upon

simultaneous improvements in theproductivity of capital and labour.

• The results of CGE studies results applysolely to energy efficiencyimprovements by producers, sotherefore cannot be extended toenergy efficiency improvements byconsumers. Also, the small number ofstudies available, the diversity ofapproaches used and themethodological weaknesses of CGEmodeling all suggest the need forcaution when interpreting theseresults.

• In principle, more robust estimates ofthe economy-wide rebound effect maybe obtained from macro-econometricmodels of national economies. Barkerand Foxon (2006) use this approach toestimate an economy-wide reboundeffect of 26% from current UK energyefficiency policies. However, there are anumber of reasons why this study mayhave underestimated the economy-wide rebound effect.

• The main insight from this evidencebase is the dependence of theeconomy-wide rebound effects on thenature and location of the energyefficiency improvement, which makesany general statements regarding themagnitude of such effectsquestionable. However, while littleconfidence can be placed in thequantitative estimates, the frequentfinding that economy-wide reboundeffects exceed 50% should give causefor concern. Extending and improvingthis evidence base should therefore bea priority for future research.

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

60

15

20 25

30

S4134 12/10/07 13:56 Page 60

This section summarises the empiricalevidence for the Khazzoom-Brookespostulate, focusing in particular on thework of W.S. Jevons, Len Brookes andHarry Saunders. An in-depth examinationof this evidence is contained in TechnicalReport 5.

The ‘Khazzoom-Brookes’ (K-B) postulateclaims that economy-wide rebound effectsexceed unity, so that energy efficiencyimprovements lead to backfire. The termwas first coined by Saunders (1992) whorefers specifically to the energy efficiencyimprovements that result from technicalchange, rather than from the substitutionof other inputs for energy (Box 2.1). Also,Saunders’s statement of the K-B postulateimplies that ‘pure’ energy efficiencyimprovements will increase energyconsumption, regardless of anyassociated improvements in theproductivity of capital, labour andmaterials. However, this distinction is not

made by other advocates of the K-Bpostulate, including Len Brookes himself(Brookes, 2000).

Neither Brookes nor Saunders citeempirical estimates of direct, indirect andeconomy-wide rebound effects, butinstead develop their case through a mixof theoretical argument, mathematicalmodelling, anecdotal examples and‘suggestive’ evidence from econometricanalysis and economic history (Box 5.1).It is these ‘indirect’ sources of evidencethat are reviewed in this section. None ofthis research provides quantitativeestimates for the size of the economy-wide rebound effect and the majority ofthe studies make no reference to therebound effect at all. Instead, theyprovide evidence that may arguably beused in support of the K-B postulate,either on theoretical grounds or on thebasis of historical experience. Like the K-B postulate itself, this evidence frequently

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

61

5. The evidence for ‘backfire’

• The relationship between macro-level energy productivity and micro-levelthermodynamic efficiency (Berndt, 1978; 1990; Ang, 2006).

• The relationship between energy productivity and total factor productivity (Schurr,1982; 1983; 1985; Jorgenson, 1996).

• The contribution of energy to economic growth (Toman and Jemelkova, 2003; Sternand Cleveland, 2004).

• The implications of energy augmenting technical change within neoclassicalproduction and growth theory (Saunders, 1992; 2007).

• Decomposition analysis of historical trends in energy consumption (Schipper andGrubb, 2000).

• Econometric studies of the ‘causal’ relationships between energy consumption andeconomic growth (Kraft and Kraft, 1978; Stern, 1993).

• Empirically validated alternatives to conventional models of economic growth thatinclude energy or useful work as a factor of production (Kummel, et al., 2002; Ayresand Warr, 2005).

Box 5.1 Sources of evidence considered indirectly relevant to the K-B postulate

S4134 12/10/07 13:56 Page 61

runs counter to conventional wisdom.However, in all cases it is suggestiverather than definitive.

This section is structured as follows:

Section 5.1 provides a historical contextto the debate, including Jevons’ 19thcentury example of the effect of energyefficiency improvements in steamengines, together with morecontemporary examples of comparable‘general-purpose’ technologies. Thisintroduces the central theme of thissection: namely that the evidence used insupport of the K-B postulate is closelylinked to a broader question regarding thecontribution of energy to economicgrowth.

Section 5.2 summarises Brookes’arguments in favour of the postulate,identifies some empirical and theoreticalweaknesses and examines whether morerecent research supports Brookes’ claims.Section 5.3 provides a non-technicalsummary of Saunders’ work, highlightingthe dependence of these results onspecific theoretical assumptions and thequestions it raises for standard economicapproaches.

Section 5.4 investigates the empiricalevidence regarding the scope forsubstitution between energy and capital.It shows how 30 years of research havefailed to reach a consensus on this issueand argues that the relationship betweenthis and the rebound effect is morecomplex than some authors havesuggested. Section 5.5 examines furtherevidence on the contribution of energy toeconomic growth and points to theinteresting parallels between Brookes’

arguments and those of contemporaryecological economists. While the evidenceremains ambiguous and open tointerpretation, the central argument isthat energy – and by implication energyefficiency - plays a significantly moreimportant role in economic growth than isassumed within mainstream economics.Section 5.6 highlights some of theimplications of this finding for theeconomy-wide rebound effect. Section 5.7concludes.

5.1 Historical perspectivesThe argument that improved energyefficiency will increase economy-wideenergy consumption was first made byW.S. Jevons in 1865, who used theexample of the steam engine:

“… it wholly a confusion of ideas tosuppose that the economical use of fuel isequivalent to a diminished consumption.The very contrary is the truth....…Everyimprovement of the engine when effectedwill only accelerate anew the consumptionof coal…” (Jevons, 1865)

Jevons cites the example of the Scottishiron industry, in which:

“..... the reduction of the consumption ofcoal, per ton of iron, to less than one thirdof its former amount, has beenfollowed….by a tenfold increase in totalconsumption, not to speak of the indirecteffect of cheap iron in accelerating othercoal consuming branches of industry…”(Jevons, 1865)

According to Jevons, the early Savoryengine for pumping floodwater out of coal

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

62

S4134 12/10/07 13:56 Page 62

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

63

mines “…consumed no coal because it rateof consumption was too high.” It was onlywith the subsequent improvements byWatt and others that steam enginesbecame widespread in coal mines,facilitating greater production of lowercost coal which in turn was used bycomparable steam engines in a host ofapplications. One important applicationwas to pump air into blast furnaces,thereby increasing the blasttemperatures, reducing the quantity ofcoal needed to make iron and reducingthe cost of iron (Ayres, 2002). Lower costiron, in turn, reduced the cost of steamengines, creating a positive feedbackcycle. It also contributed to thedevelopment of railways, which loweredthe cost of transporting coal and iron,thereby increasing demand for both.

Jevons highlighted the fact thatimprovements in the thermodynamicefficiency of steam engines wereintertwined with broader technicalchange, including: “…. contrivances, suchas the crank, the governor, and the minormechanism of an engine, necessary forregulating, transmitting, or modifying itspower….” (Jevons, 1865).41 Thesedevelopments were essential to theincreased use of steam engines as asource of motive power and demonstratehow improvements in thermodynamicefficiency are frequently linked to broaderimprovements in technology and totalfactor productivity.

More recently, Rosenberg (1989) has citedthe comparable example of the Bessemerprocess for steelmaking:

“[the Bessemer process] was one of themost fuel saving innovations in the historyof metallurgy [but] made it possible toemploy steel in a wide variety of uses thatwere not feasible before Bessemer,bringing with it large increases indemand. As a result, although the processsharply reduced fuel requirements perunit of output, its ultimate effect was toincrease....the demand for fuel.”(Rosenberg, 1989)

The low cost Bessemer steel initially founda large market in the production of steelrails, thereby facilitating the growth of therail industry, and later in a much widerrange of applications includingautomobiles. However, the mild steelproduced by the Bessemer process is avery different product to wrought iron(which has a high carbon content) and issuitable for a much wider range ofapplications. Hence, once again, theimprovements in the thermodynamicefficiency of production processes aredeeply entwined with broaderdevelopments in process and producttechnology.

These examples relate to energyefficiency improvements in the earlystages of development of energy intensiveprocess technologies and intermediategoods that have the potential forwidespread use in multiple applications. It

41 Another factor was the development of high pressure boilers after 1840 (Hills, 1989). Compared to the early Newcomenengine, the Watt engine used only 2/5 of the amount of coal per unit of output, while compared to the early Watt engine, highpressure steam turbines in the earlier 20th-century used only one sixth (Crafys, 2003). According to Atack (1979), the annualcost of a horsepower of steam fell by just over 80% between the 1820s and the 1890s.

S4134 12/10/07 13:56 Page 63

is possible that the same consequencesmay not follow for energy efficiencyimprovements in mature and/or non-energy intensive process technologies andintermediate goods that have a relativelynarrow range of applications. Similarly,the same consequences may not followfrom improvements in consumertechnologies that supply energy serviceswith a low own-price elasticity and whereenergy represents only a small share oftotal costs. (Ayres, 2002). The relevanceand importance of these distinctions isdiscussed further below.

A historical perspective on rebound effectsis provided by Fouquet and Pearson(2006), who present some remarkabledata on the price and consumption oflighting services in the UK over a period ofseven centuries (Table 5.1). Per capitaconsumption of lighting services grewmuch faster than per capita GDPthroughout this period, owing in part tocontinuing reductions in the price oflighting services (£/lumen hour). This, inturn, derived from continuingimprovements in the thermodynamicefficiency of lighting technology, incombination with continuing reductions inthe real price of lighting fuel (itself, partlya consequence of improvements in thethermodynamic efficiency of energysupply). In this case, improvements inlighting technology were substantiallymore important than improvements inenergy supply (in the ratio of 180 to 1over the period 1800 to 2000).

Per capita lighting consumption increasedby a factor of 6566 between 1800 and2000, largely as a consequence of thefalling cost of lighting services relative toincome, but also as a result of the boost

to per capita GDP provided by thetechnical improvements in lightingtechnology. Since lighting efficiencyimproved by a factor of 1000, the datasuggest that per capita energyconsumption for lighting increased by afactor of six. In principle, the directrebound effect could be estimated byconstructing a counterfactual scenario inwhich lighting efficiencies remained at1800 levels. But this would be ameaningless exercise over such a timeinterval, given the co-evolution andinterdependence of the relevant variablesand the interrelationship between energyconsumption and economic growth (Box5.2). To the extent that the demand forlighting is approaching saturation in manyOECD countries, future improvements inlighting efficiency may be associated withsmaller rebound effects. Nevertheless,this historical perspective gives cause forconcern over the potential of technologiessuch as compact fluorescents to reduceenergy consumption in developingcountries.

5.2 Energy productivity andeconomic growth Despite their far-reaching implications,Jevons’ ideas were neglected untilcomparatively recently and contemporaryadvocates of energy efficiency arefrequently unaware of them. While a 1980paper by Daniel Khazzoom stimulatedmuch research and debate on the directrebound effect (Besen and Johnson,1982; Einhorn, 1982; Henly, et al., 1988;Lovins, et al., 1988), most researchersignored the long-term, macroeconomic

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

64

S4134 12/10/07 13:56 Page 64

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

65

Year Price of Lighting Price of Consumption Total Reallighting efficiency lighting of light per consumption GDP per

fuel services capita of light capita

1300 1.50 0.50 3.0 - - 0.25

1700 1.50 0.75 2.0 0.17 0.1 0.75

1750 1.65 0.79 2.1 0.22 0.15 0.83

1800 1.0 1 1 1 1 1

1850 0.40 4.4 0.27 3.9 7 1.17

1900 0.26 14.5 0.042 84.7 220 2.9

1950 0.40 340 0.002 1528 5000 3.92

2000 0.18 1000 0.0003 6566 25630 15

Table 5.1 Seven centuries of lighting in the UK

Note: 1800=1.0 for all indices Source: Fouquet and Pearson (2006)

Numerous studies have demonstrated strong correlations between economic output and energyconsumption at different levels of aggregation and over different periods of time. But this leavesopen the question of to what extent the growth in economic output can be considered a cause ofthe increased energy consumption, and to what extent the growth in energy inputs can beconsidered a cause of the increased economic output. Alternatively there could be a synergisticrelationship between the two, with each causing the other as part of a positive feedbackmechanism (Ayres and Warr, 2002).

The conventional wisdom (as represented by both neoclassical and ‘endogenous’ growth theory)is that increases in energy inputs play a relatively minor role in economic growth, largely becauseenergy accounts for a relatively small share of total costs (Jones, 1975; Denison, 1985; Gullicksonand Harper, 1987; Barro and Sala-I-Martin, 1995). Economic growth is assumed to result insteadfrom the combination of increased capital and labour inputs, changes in the quality of those inputsand increases in total factor productivity that are frequently referred to as technical change.However, this view has been contested by ecological economists, who argue that the increasedavailability of high quality energy inputs has been the primary driver of economic growth over thelast two centuries (Cleveland, et al., 1984; Beaudreau, 1998; Kummel, et al., 2000). According tothis view, the productivity of energy inputs is substantially greater than is suggested by its smallshare of total costs.

The conventional and ecological perspectives reflect differing assumptions and are supported byconflicting empirical evidence. A difficulty with both is that they confine attention to therelationship between energy consumption and economic growth. But the reason that energy iseconomically significant is that it is used to perform useful work - either in the form of mechanicalwork (including electricity generation) or in the production of heat (Ayres and Warr, 2005). Moreuseful work can be obtained with the same, or less, energy consumption through improved energyefficiency. Hence, if increases in energy inputs contribute disproportionately to economic growth,then improvements in energy efficiency may do the same. Conversely, if increases in energyinputs contribute little to economic growth, then neither should improvements in energyefficiency. This suggests that differing views over the size of the economy-wide rebound effectmay partly reflect differing views over the relationship between energy and economic growth.

Box 5.2 Energy consumption and economic growth - neoclassical and ecological perspectives

S4134 12/10/07 13:56 Page 65

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

66

implications that were Jevons’ primaryconcern. However, Jevons’ argumentshave been taken up with some vigour bythe British economist, Len Brookes, whohas developed coherent arguments infavour of the K-B postulate and combinedthese with critiques of government energyefficiency policy (Brookes, 1978; 1984;1990; 2000; 2004). Brookes work hasprompted a fierce response from critics(Grubb, 1990; Herring and Elliot, 1990;Toke, 1990; Grubb, 1992), to whichBrookes has provided a number of robustresponses (Brookes, 1992; 1993).42

Brookes (2000) argues that “…The claimsof what might be called the Jevons schoolare susceptible only to suggestiveempirical support”, since estimating themacroeconomic consequences ofindividual improvements in energyefficiency is practically impossible. Hetherefore relies largely on theoreticalarguments, supported by indirect sourcesof evidence, such as historical correlationsbetween various measures of energyefficiency, total factor productivity,economic output and energy consumption(Schurr, 1984; 1985). A key argumentruns as follows:

“…it has been claimed since the time ofJevons (1865) that the market for a moreproductive fuel is greater than for lessproductive fuel, or alternatively that for aresource to find itself in a world of moreefficient use is for it to enjoy a reductionin its implicit price with the obviousimplications for demand.”

However, Brookes’ use of the term

‘implicit price’ is confusing. Individualenergy efficiency improvements do notchange the price of input energy, butinstead lower the effective price of outputenergy, or useful work. For example,motor-fuel prices may be unchangedfollowing an improvement in vehicle fuelefficiency, but the price per vehiclekilometre is reduced. The ‘obviousimplications’ therefore relate to thedemand for useful work, and not to thedemand for energy commoditiesthemselves. While the former may beexpected to increase, energy demandmay either increase or decreasedepending upon the price elasticity ofdemand for useful work and theassociated indirect rebound effects.

Of course, the combined impact ofmultiple energy efficiency improvementscould lower energy demand sufficiently toreduce energy prices and therebystimulate a corresponding increase ineconomy-wide energy demand. Thisforms one component of the economy-wide rebound effect. But while it isobvious that the overall reduction inenergy consumption will be less thanmicroeconomic analysis suggests, thistheoretical argument appear to be aninsufficient basis for claiming that backfireis inevitable.

Brookes also criticises the assumptionthat energy service demand will remainfixed while the marginal cost of energyservices falls under the influence of raisedenergy efficiency, and the relatedassumption that individual energy savingscan be added together to produce an

42 One of these critics, Horace Herring, is now more sympathetic to the K-B postulate (Herring, 2006)

S4134 12/10/07 13:56 Page 66

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

67

estimate of what can be saved over theeconomy as a whole. In both cases,Brookes is highlighting the persistentneglect of both direct and indirectrebound effects in the conventionalassessment of energy efficiencyopportunities. However, arguing that theeconomy-wide rebound effect is greaterthan zero is different from arguing that itis greater than one – as the K-B postulatesuggests.

Brookes marshals a number of otherarguments in support of the K-B postulatethat appear more amenable to empiricaltest. In doing so, he highlights someimportant issues regarding therelationship between energy consumptionand economic growth. The three mostimportant arguments may becharacterised as follows:

• The productivity argument: Theincreased use of higher quality formsof energy (especially electricity) hasencouraged technical change,substantially improved total factorproductivity and driven economicgrowth. Despite the substitution ofenergy for other inputs, this technicalchange has stimulated a sufficientlyrapid growth in economic output thataggregate energy efficiency hasimproved at the same time asaggregate energy consumption hasincreased.43 This pattern may beexpected to continue in the future.

The productivity argument rests upontwo separate, but related sources ofempirical evidence. First, the work ofSam Schurr and colleagues on thehistorical importance of changes inenergy quality (notably electrification)in driving productivity growth (Schurr,et al., 1960; Schurr, 1984; 1985).Second, the work of Jorgenson andothers on the direction of technicalchange (Jorgenson, 1984; Hogan andJorgenson, 1991). Contrary tostandard assumptions, Jorgenson’sresults suggest that, at the level ofindividual sectors, the contribution oftechnical change has been to increaseenergy intensity over time, ratherthan reduce it.44 This work is alsocited as suggestive evidence for the K-B postulate by Saunders (1992).

• The endogeneity argument: Acommon approach to quantifying the‘energy savings’ from energyefficiency improvements is to holdenergy intensity fixed at some historicvalue and estimate what consumption‘would have been’ in the absence ofthose improvements (Geller, et al.,2006). The energy savings fromenergy efficiency improvements arethen taken to be the differencebetween the actual demand and thecounterfactual scenario. But if theenergy efficiency improvements are anecessary condition for the growthin economic output, the construction

43Between 1918 and 1973, the US economy experienced substantial increases in both total factor productivity and energyefficiency, while energy prices fell in real terms. Even though energy use increased relative to labour and capital, the increaseduse of more flexible forms of energy (oil and electricity) enhanced the productivity of capital and labour sufficiently that energyuse per unit of output fell (Schurr, 1982).

44 In their original study, Jorgenson and Fraumeni (1981) estimated the rate of change in the share of energy costs in the valueof output of US manufacturing sectors, holding input prices constant. Contrary to conventional wisdom, Jorgenson and Fraumenifound ‘energy-using’ technical change in 29 out of 35 sectors. Generally, we would expect energy saving (using) technical changeto be associated with falling (increasing) energy intensity. However, this may not always be the case, since it also depends uponthe rate of change in total factor productivity (Sanstad, et al., 2006). With energy-using technical change a decrease (increase)in the price of energy will raise (lower) total factor productivity.

S4134 12/10/07 13:56 Page 67

of a counterfactual in this way ismisconceived.45 The endogeneityargument is not developed in detail byBrookes, but raises questions over theuse of ‘decomposition analysis’46 toexplore the rebound effect (Schipperand Grubb, 2000).

• The accommodation argument:Energy efficiency improvements areclaimed to ‘accommodate’ an energyprice shock so that the energysupply/demand balance is struck at ahigher level than if energy efficiencyhad remained unchanged (Brookes,1984). While not immediatelyobvious, this argument rests on theassumption that the income elasticityof ‘ useful’ energy demand fallssteadily as an economy develops, butis always greater than unity (Brookes,1972). ‘Useful’ energy consumption isa quality-adjusted measure ofaggregate energy consumption inwhich different energy types areweighted by their relative economicproductivities (Adams and Miovic,1968).

Technical Report 5 describes the historicalresearch that forms the basis for thesearguments, summarises how Brookesuses this research to support his case,identifies potential empirical andtheoretical weaknesses and examines indetail whether more recent researchconfirms or contradicts Brookes’ claims. Itpoints to a number of flaws, both in theevidence itself and in the manner in which

Brookes uses this evidence to support hiscase. Specific criticisms include thefollowing:

• Schurr’s work applies primarily to thecausal effect of shifts to higher qualityfuels (notably electricity), rather thanimprovements in thermodynamicconversion efficiency or other factorsthat affect aggregate measures ofenergy efficiency. The effect of thelatter on total factor productivity maynot be the same as the effect of theformer. Also, the patterns Schurruncovered may not be as ‘normal’ asBrookes suggests and the linkbetween energy efficiencyimprovements and improvements intotal factor productivity appears tovary greatly, both over time andbetween different sectors and energyservices.

• Neither Jorgenson’s work itself, northose of comparable studiesconsistently find technical change tobe ‘energy-using’. Instead, theempirical results vary widely betweendifferent sectors, countries and timeperiods and are sensitive to minorchanges in econometric specification(Norsworthy, et al., 1979; Roy, et al.,1999; Welsch and Ochsen, 2005;Sanstad, et al., 2006). Jorgenson’sresults rest on the erroneousassumption that the rate and directionof technical change is fixed, and moresophisticated models suggest that themagnitude and sign of technical

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

68

45 Technically, economic output and energy efficiency should be considered endogenous variables, since each influences theother. Conventional methodological approaches to estimating historical ‘energy-savings’, such as decomposition analysis, fail todo this.

46 Decomposition analysis expresses trends in aggregate quantities as the product of a number of different variables. Forexample, economy-wide energy consumption may be expressed into the product of population, GDP per capita and energy useper unit of GDP. An additive decomposition expresses the change in energy use over a particular period as the sum of the changein each of the right-hand side variables, while a multiplicative decomposition expresses the ratio of energy use at the end of theperiod to that at the beginning of the period as the product of comparable ratios for each of the right-hand side variables. Similarexpressions can be developed at varying levels of detail for energy use within individual sectors. Thanks in part to the work ofLee Schipper and colleagues, decomposition analysis has become a widely used tool within energy economics (Ang, 1999).

S4134 12/10/07 13:56 Page 68

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

69

change varies between sectors andtypes of capital as well as over time(Sue Wing and Eckaus, 2006). Also,the failure to check for the presence of‘cointegration’47 in the data or toaccount for changes in energy qualitymeans that some of the estimatescould be either biased or spurious(Kaufmann, 2004). Moreover, even ifenergy-using technical change wereto be consistently found, therelationship between this finding andthe K-B postulate remains unclear.48

• The endogeneity argument isrhetorically persuasive but lacks afirm empirical basis. The relativeimportance of energy efficiencyimprovements (however defined)compared to other forms of technicalchange in encouraging economicgrowth remains to be established.

• The ‘accommodation’ argument isbased upon a highly simplifiedtheoretical model of the worldeconomy (Brookes, 1984), which isboth unconventional in approach anddifficult to interpret and calibrate. Themodel rests on the assumption thatthe income elasticity of ‘useful’ energydemand is always greater than unity,thereby allowing economic output tobe represented as a linear function ofuseful energy inputs. An earlier study

by Brookes (1972) provides somesupport for this hypothesis, but thishas not been updated. Contemporaryresearch on ‘Environmental KuznetsCurves’ has generally not tested thishypothesis, since it aggregatesenergy consumption on the basis ofthermal content (Stern, 2004c;Richmond and Kaufmann, 2006b;2006a). However, analysis by Dargay(1992) suggests that the incomeelasticity of useful energy demand hasdeclined from above unity in the1960s to around 0.7 in 1990, whichcontradicts Brookes’ claims.

In sum, not only does each of the sourcesof evidence have empirical and theoreticalweaknesses, but the extent to which they(individually and collectively) support theK-B postulate is open to question. Hence,while Brookes has highlighted someimportant issues and pointed to sources ofevidence that challenge conventionalwisdom, he has not provided a convincingcase in support of the K-B postulate.

Perhaps the most important insight fromBrookes’ work is that improvements inenergy productivity are frequentlyassociated with improvements in theproductivity of capital, labour andmaterials. In particular, historicalexperience suggests that improvements in

47 The presence of cointegration between two or more variables implies that one variable cannot move ‘too far’ away fromanother, because there is a long-term relationship between them. This may be because one variable ‘causes’ the other, or thatthey are both driven by a third, possibly omitted, variable. Cointegration analysis seeks to identify this relationship by detectingwhether the irregular trends in a group of variables are shared by the group, so that the total number of unique trends is lessthan the number of variables. Failing to allow for cointegration can allow the precision of relationships to be overestimated andmay also lead to incorrect signs (Lim and Shumway, 1997).

48 Jorgenson’s work suggests that the contribution of technical change has frequently been to reduce energy efficiency andthereby increase overall energy consumption, even while other factors (such as structural change) are acting to decrease it.Hence, not only is the direction of technical change opposite to what is conventionally assumed, but also opposite to what isrequired for an empirical estimate of the rebound effect. But technical change has clearly improved the thermodynamicconversion efficiency of individual devices, such as motors and boilers. What Jorgenson’s work suggests, therefore, is this thathas not necessarily translated into improvements in more aggregate measures of energy intensity at the level of industrialsectors. Similarly, the more robust results of Sue Wing and Eckhaus (2006) suggest that this has not necessarily translated intoimprovements in more aggregate measures of energy intensity for particular types of capital (e.g. machinery). Hence, therelevance of these results may hinge upon the appropriate choice of independent variable for the rebound effect.

S4134 12/10/07 13:56 Page 69

energy productivity have been associatedwith proportionally greater improvementsin total factor productivity. While Schurr’swork provides evidence for this at thelevel of the national economy, numerousexamples from the energy efficiencyliterature provide comparable evidence atthe level of individual sectors andtechnologies (Box 5.3). Such examplesare frequently used by authors such asLovins (1997) to support the businesscase for energy efficiency, but they alsopoint to situations where rebound effectsmay be expected to be large (Saunders,2000b). If energy efficient technologiesboost total factor productivity and thereby

save more than energy costs alone, theargument that rebound effects must besmall because the share of energy in totalcosts is small is undermined. Much thesame applies to the contribution of energyefficiency improvements to economicgrowth. But this leaves open the questionof whether energy efficiencyimprovements are necessarily associatedwith proportionally greater improvementsin total factor productivity, or whether (asseems more likely) this is contingent uponparticular technologies andcircumstances. If the latter is the case,policy measures could potentially betargeted on ‘dedicated’ energy efficiency

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

70

• Lovins and Lovins (1997) used case studies to argue that better visual, acoustic andthermal comfort in well-designed, energy efficient buildings can improve labourproductivity by as much as 16%. Since labour costs in commercial buildings aretypically 25 times greater than energy costs, the resulting cost savings canpotentially dwarf those from reduced energy consumption.

• Pye and McKane (1998) showed how the installation of energy efficient motorsreduced wear and tear, extended the lifetime of system components and achievedsavings in capital and labour costs that exceeded the reduction in energy costs.

• Sorrell et al. (2004) found a host of examples of the ‘hidden benefits’ of energyefficiency improvements within 48 case studies of organisational energymanagement. For example changes to defrosting regimes at a brewery led to energysavings, water savings, reduced maintenance and reduced deterioration of buildingfabric.

• Worrell et al. (2003) analysed the cost savings from 52 energy efficiency projects,including motor replacements, fans/duct/pipe insulation, improved controls and heatrecovery in a range of industrial sectors. The average payback period from energysavings alone was 4.2 years, but this fell to 1.9 years when the non-energy benefitswere taken into account.

• Using plant-level data, Boyd and Pang (2000) estimated fuel and electricity intensityin the glass industry as a function of energy prices, cumulative output, a time trend,capacity utilisation and overall productivity. Their results show that the mostproductive plants are also most energy efficient and that a 1% improvement inoverall productivity results in a more than 1% improvement in energy efficiency.

Box 5.3 Examples from the energy efficiency literature of the link betweenimproved energy efficiency and improved total factor productivity

S4134 12/10/07 13:56 Page 70

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

71

improvements that have smaller impactson total factor productivity and hencesmaller rebound effects.

5.3 Energy productivity andproduction theory Harry Saunders is the second majoradvocate of the ‘Khazzoom-Brookespostulate’ and has brought a new level ofsophistication to the rebound debate bybasing his arguments upon neoclassicalproduction and growth theory. This workis abstract and theoretical and isnecessarily based on highly restrictiveassumptions that can be a focus ofcriticism. But Saunders does not claimthat his results prove the K-B postulate,but merely provide suggestive evidence inits favour, given certain assumptionsabout how the economy operates. Mostimportantly, he shows how backfire is thepredicted outcome of standard economicmodels.

Saunders (1992) uses the neoclassicalgrowth model to argue that backfire is alikely outcome of ‘pure’ energy efficiencyimprovements - that is, a form oftechnical change that improves energyproductivity while not affecting theproductivity of other inputs. He alsoargues that improvements in capital,labour or materials productivity willincrease overall energy consumption.Since technical change typically improvesthe productivity of several inputs

simultaneously, Saunders argues thatmost forms of technical change willincrease overall energy consumption.

Saunders’ use of the neoclassical growthmodel was subsequently challenged byHowarth (1997), who argued that thefailure to distinguish between energy andenergy services led to the probability ofbackfire being overestimated. However,Saunders (2000a) subsequentlydemonstrated that backfire is stillpredicted by the neoclassical model whenan alternative choice is made for theproduction function49 used to provideenergy services. In a more recentcontribution, Saunders (2007) focuses onthe potential of different types ofproduction function to generate backfire.Unlike Saunders (1992), this work is alsoapplicable to individual firms and sectorsand opens up the possibility of usingempirically estimated production functionsto estimate the rebound effect fromparticular technologies in particularsectors (Saunders, 2005). Box 5.4provides a very simple illustration ofSaunders’ approach and contrasts thiswith an alternative approach by Laitner(2000).

Saunders (2007) shows how the predictedmagnitude of rebound effects dependsalmost entirely on the choice of therelevant production function – whether atthe firm, sector or economy-wide level.Several commonly used productionfunctions are found to be effectivelyuseless in investigating the rebound

49 Production functions are normally represented by functional forms which are intended to approximate the relationshipbetween inputs and outputs. Their purpose is to define a set of parameters which, given the observed data, reasonablyapproximate real-world production behaviour. This is useful in understanding the rate at which agents within an economy canfeasibly (or have historically) moved between alternative input combinations. There are a range of different functional formsavailable and the appropriate choice involves trade-offs between flexibility and analytical tractability.

S4134 12/10/07 13:56 Page 71

effect, since the relevant results are thesame for whatever values are chosen forkey parameters. One popular productionfunction (the constant elasticity ofsubstitution, or CES), is found to be ableto simulate rebound effects of differentmagnitudes, but only if a particularassumption is made about how differentinputs are combined.50 Since this form iswidely employed within energy-economicmodels, Saunders’ results raise seriousconcerns about the ability of such modelsto accurately simulate rebound effects. Analternative and more flexible functionalform (the ‘translog’) that is widely used inempirical studies is also found to lead tobackfire once standard restrictions areimposed on the parameter values toensure that the behaviour of the functionis consistent with economic theory(Saunders, 2007)51.

There is a substantial empirical literatureestimating the parameters of differenttypes of production function at differentlevels of aggregation and obtaining a goodfit with observed data. Hence, if suchfunctions are considered to provide areasonable representation of real-worldeconomic behaviour, Saunders’ worksuggests that ‘pure’ energy-efficiency

improvements are likely to lead tobackfire. Alternatively, if rebound effectsare considered to vary widely inmagnitude between different sectors,Saunders’ work suggests that standardand widely used economic methodologiescannot be used to simulate them.

The above conclusions apply to pureenergy efficiency improvements. ButSaunders (2005) also uses numericalsimulations to demonstrate the potentialfor much larger rebound effects whenimprovements in energy efficiency arecombined with improvements in theproductivity of other inputs. Again, if thevalidity of the theoretical assumptions isaccepted, these results suggest thatbackfire may be a more common outcomethan is conventionally assumed.

Saunders approach is entirely theoreticaland therefore severely limited by theassumptions implicit in the relevantmodels. For instance, technology alwayscomes free, there are only constantreturns to scale in production, marketsare fully competitive, there is always fullemployment, qualitative differences incapital and energy are ignored and so on.Indeed, a considerable literaturechallenges the idea that an ‘aggregate’

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

72

50 The CES function used by Saunders combines inputs into pairs, or ‘nests’. For example, a nested production function withcapital (K), labour (L) and energy (E) inputs, could take one of three forms, namely: K(LE); (KL)E; (KE)L. Saunders (1992) showsthat the (KL)E form permits a range of values for the rebound effect (depending upon the elasticity of substitution betweenenergy and capital/labour), while the other forms always lead to backfire. However, despite being in widespread use, this typeof function imposes very restrictive conditions on real-world behaviour that are not supported by empirical evidence (Frondel andSchmidt, 2004).

51 Restrictions normally have to be imposed upon the parameter values in a translog cost function to ensure that its behaviouris consistent with basic economic theory. In particular, the cost function must be concave - implying that the marginal productof each input declines with increasing use of that input. In many applications, such as CGE modelling, these conditions need tobe satisfied for all input combinations, but empirically estimated cost functions sometimes violate these conditions (Diewert andWales, 1987). However, Ryan and Wales (2000) show that if concavity is imposed locally at a suitably chosen reference point,the restriction may be satisfied at most of the data points in the sample. Under these circumstances, the translog may be ableto represent different types of rebound effect for particular data sets.

S4134 12/10/07 13:56 Page 72

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

73

production function for the economy as awhole is meaningful concept (Fisher,1993; Temple, 2006) - although this maynot necessarily invalidate the use of suchfunctions for representing the behaviourof individual sectors. A particularweakness is the assumption that technicalchange is costless and autonomous,without explicit representation of theprocesses that affect its rate anddirection. This characteristic limits thecapacity of such models to address manypolicy-relevant questions. More recentdevelopments in so-called ‘endogenousgrowth theory’ have overcome thisweakness to a large extent, but to date noauthors have used such models to explorethe rebound effect. However, since what isat issue is the consequences of energyefficiency improvements, the source ofthose improvements is arguably asecondary concern.

Overall, Saunders work suggests thatsignificant rebound effects can exist intheory, backfire is quite likely and thisresult is robust to different modelassumptions. Since these results derivefrom a contested theoretical framework,they are suggestive rather than definitive.But they deserve to be taken seriously.

5.4 Substitution betweenenergy and capitalA key conclusion from Saunders’ work isas follows:

“It appears that the ease with which fuelcan substitute for other factors ofproduction (such as capital and labour)has a strong influence on how much

rebound will be experienced. Apparently,the greater this ease of substitution, thegreater will be the rebound” (Saunders,2000, p. 443).

The parameter that measures this ‘ease’of substitution is the so-called elasticity ofsubstitution between energy and otherinputs (σ). High values of the elasticity ofsubstitution between energy and otherinputs mean that a particular sector oreconomy is more ‘flexible’ and maytherefore adapt relatively easily tochanges in energy prices. In contrast, lowvalues of the elasticity of substitutionbetween energy and other inputs suggestthat increases in energy prices may havea disproportionate impact on productivityand growth. The elasticity of substitutionis therefore a key parameter withinenergy-economic models, leadingSaunders to suggest a possible trade off inclimate policy:

“…If one believes σ is low, one worries lessabout rebound and should incline towardsprogrammes aimed at creating new fuelefficient technologies. With low σ carbontaxes are less effective in achieving agiven reduction in fuel use and wouldprove more costly to the economy. Incontrast, if one believes σ is high, oneworries more about rebound and shouldincline towards programmes aimed atreducing fuel use via taxes. With high σ,carbon taxes have more of an effect atlower cost to the economy.” (Saunders,2000b)

These observations suggest that a closerexamination of the nature, determinantsand typical values of elasticities ofsubstitution between energy and otherinputs could provide some insights into

S4134 12/10/07 13:56 Page 73

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

74

Much of Saunders work involves the use of calculus to investigate the behaviour ofneoclassical production functions. A very simple, but still widely used production functionis the ‘Cobb- Douglas’, which may be represented as follows:

Where: K=capital, L=labour, E=energy and α+β=1. The multiplier τ(τ≥1) increases theproductivity of energy inputs, so that the product τE represents ‘effective’ energy inputs.This may be interpreted as a form of technical change that does not affect theproductivity of other inputs and is assumed to be costless. Saunders (2007) investigatesthe effect of improvements in energy productivity (τ) on energy consumption andeconomic output in the short-term, with real energy prices fixed. He finds that, with aCobb-Douglas function, both energy consumption and output increase by the sameamount, leaving the aggregate energy/output ratio unchanged. The improvement inthe productivity of energy inputs (τ) has therefore led to backfire, since economy-wideenergy consumption has increased. Using standard assumptions about the share ofenergy in total costs, Saunders (2000a) estimates that a 20% increase in theproductivity of energy inputs should only increase GDP by some 2.3%.

This approach can be compared with Laitner’s (2000) ‘back of the envelope’ estimate ofthe effect of energy efficiency improvements on the US economy. This simply assumesthat energy efficiency policies would reduce the economy-wide energy/GDP ratio by30%. As a result, the only rebound effects that Laitner considers are the effect of energyefficiency improvements on GDP growth. Since the share of energy in total costs issmall, the additional economic growth stimulated by ‘pure’ energy efficiencyimprovements should be insufficient to offset the energy savings achieved by thereduction in the energy/GDP ratio. Laitner shows that this conclusion is unchanged whenpossible changes in energy prices are also taken into account.

Both authors agree, therefore, that ‘pure’ energy efficiency improvements should havea relatively small impact on GDP. But while Laitner uses this result to argue that backfireis unlikely, Saunders position is that backfire is likely to be the norm.

These conflicting conclusions result in part from differences in approach, and in partfrom different definitions of the independent variable. Saunders uses an improvement inenergy productivity as the independent variable and derives a result in which theeconomy-wide energy/GDP ratio remains unchanged (i.e. energy consumption andeconomic output increase by the same amount). Laitner, in contrast, uses the economy-wide energy/GDP ratio as the independent variable and simply assumes that energyefficiency policies will reduce this by 30% - thereby ignoring any ‘lower-level’ reboundeffects from these policies. A criticism of Saunders approach could be that the real-worldeconomy is unlikely to behave in the manner suggested by a Cobb-Douglas productionfunction. But Saunders (2007) shows that several alternative and more flexibleassumptions about the form of the production function also lead to backfire. In contrast,a criticism of Laitner’s approach could be that the assumption that energy efficiencypolicies will reduce the energy/GDP ratio by this amount is flawed.

Hence, while both approaches are internally consistent, their results are driven byconflicting theoretical assumptions that require empirical validation.

Box 5.4 How theoretical assumptions can determine results

S4134 12/10/07 13:56 Page 74

the likely magnitude of rebound effects indifferent sectors and for the economy as awhole. Of particular interest is theelasticity of substitution between energyand capital, since many types of energyefficiency improvement may beunderstood as the substitution of capitalfor energy. Technical Report 3 thereforeprovides an in-depth examination ofempirical estimates of the elasticity ofsubstitution between energy and capital,together with an extensive review of theassociated theoretical issues. Theempirical literature on this subject turnsout to be confusing and contradictory,with more than three decades of empiricalresearch failing to reach a consensus onwhether energy and capital may beconsidered as ‘substitutes’ or‘complements’ (Box 5.5 and 5.6).Moreover, the relationship between theelasticity of substitution and the reboundeffect turns out to be far fromstraightforward.

Our investigation of this topic reveals thatSaunders’ statements regarding therelationship between elasticities ofsubstitution and the rebound effect arepotentially misleading. The relationshipdepends very much upon the distinctionbetween energy and energy services, theparticular choice of production function,the appropriate definition of the elasticityof substitution and the validity of theassumption of ‘separability’ betweenenergy services and other inputs (Box5.5). When these are taken into account,it is found that large rebound effects mayoccur even when the elasticity ofsubstitution between energy and capital islow, which appears to contradict theobservation quoted at the beginning of

this section. The possible trade-off inclimate policy that is suggested bySaunders may therefore not exist. Inaddition, since most empirical studiesmeasure something quite different fromthe parameters assumed within energy-economic models, the empirical basis forthose models is further called intoquestion (Box 4.4).

To the extent that a general conclusioncan be drawn from the empiricalliterature, it is that energy and capitaltypically appear to be either complementsor weak substitutes (Box 5.6). This couldhave some important implications. First, areduction in the price of capital relative toenergy (e.g. through investmentsubsidies) may in some circumstancesincrease energy consumption rather thanreduce it (Berndt and Wood, 1979).Second, the economic impact of anincrease in energy prices could besignificant:

“A reduction in the use of energy by itselfwill have a relatively small economicimpact, determined to first order byenergy’s small value share. But if thereduced use of energy also produces areduction in the use of capital, the largervalue share of capital applies and theeconomic impact is magnified. Thisindirect effect through capital can be thelargest component of the economicimpact of reduced energy use... but thiseffect is often ignored in economic impactanalyses of energy policy” (Hogan, 1979)

Hence, this review suggests the possibilityof a strong link between energyconsumption and economic output as wellas potentially high costs associated withreducing energy consumption. However,

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

75

S4134 12/10/07 13:56 Page 75

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

76

There are at least five definitions of the elasticity of substitution in common use and severalothers that appear less frequently. The lack of consistency in the use of these definitions andthe lack of clarity in the relationship between them, combine to make the empirical literatureboth confusing and contradictory (Stern, 2004a).

For all definitions, substitution between two inputs is ‘easier’ when the magnitude of theelasticity of substitution between them is greater, while the sign of the elasticity of substitutionis commonly used to classify inputs as ‘substitutes’ or ‘complements’. But the appropriateclassification depends upon the particular definition being used (i.e. inputs may be substitutesunder one measure and complements under another).

The majority of existing empirical studies use the sign of the ‘Allen-Urzwa’ elasticity ofsubstitution (AES) to make this classification. With this definition, two inputs are described assubstitutes (complements) when the usage of one increases (decreases) when the price of theother increases (decreases), holding output constant. However this measure has a number ofacknowledged drawbacks and its quantitative value lacks meaning (Frondel, 2004). In manycases, the Cross Price elasticity (CPE) or the Morishima elasticity of substitution (MES) wouldbe more appropriate measures, but these have yet to gain widespread use.

A large number of empirical studies estimate elasticities of substitution between differentinputs within different sectors and countries and over different time periods. These rely upona variety of assumptions, including in particular the specific form of the production or costfunction employed. In general, the actual scope for substitution may be expected to varywidely between different sectors, different levels of aggregation and different periods of time,while the estimated scope for substitution may depend very much upon the particularmethodology and assumptions used.

Standard methodological approaches frequently assume that the ease of substitution betweentwo inputs is unaffected by the level or price of other inputs (‘separability’). This assumptionis not always tested, and even if it is found to hold, the associated estimates of the elasticityof substitution between two inputs in the same group could still be biased (Frondel andSchmidt, 2004). Assumptions about the nature and bias of technical change may also have asubstantial impact on the empirical results, but distinguishing between technical change andprice-induced substitution is empirically challenging.

The level of aggregation of the study is also important, since a sector may still exhibit inputsubstitution in the aggregate due to changes in product mix, even if the mix of inputs requiredto produce a particular product is relatively fixed. Many studies overlook such changes andimplicitly assume that the product mix is fixed (Miller, 1986). Since the scope for changingproduct mix is greater at higher levels of aggregation, the estimated scope for inputsubstitution may also be greater when higher levels of aggregation are used (Solow, 1987).However, individual inputs cannot always be considered as independent, notably becauseenergy is required for the provision of labour and capital. Substitution of capital for energy andone sector, for example, may lead to an increase in energy consumption in the sector providingthe relevant capital. This suggests that the estimated scope for input substitution may besmaller when higher levels of aggregation are used. These two factors may therefore partlycancel each other out.

Box 5.5 Defining and measuring elasticities of substitution

S4134 12/10/07 13:56 Page 76

Technical Report 3 reviews more than 200 empirical estimates of the elasticity ofsubstitution between energy and capital. The results were analysed to see how theestimates varied with factors such as the sectors covered, the functional form employed,the use of time-series versus cross-sectional data and the assumptions regardingtechnical change. The most striking result from the analysis is the lack of consensus thathas been achieved to date, despite three decades of empirical work. While this may beexpected if the degree of substitutability depends upon the sector, level of aggregationand time period analysed, it is notable that several studies reach different conclusionsfor the same sector and time period, or for the same sector in different countries.

If a general conclusion can be drawn, it is that energy and capital typically appear to beeither complements (i.e. AES<0) or weak substitutes (i.e. 0<AES<0.5). However, littleconfidence can be placed in this conclusion, given the diversity of the results and theirapparent dependence upon the particular specification and assumptions used. Whilethere appears to be some agreement on the possible causes of the different results,there is no real consensus on either the relative importance of different causes or thelikely direction of influence of each individual cause (i.e. whether a particularspecification/assumption is likely to make the estimate of the substitution elasticitybigger or smaller).

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

77

Overall summary of results

Box 5.6 Empirical estimates of the elasticity of substitutionbetween energy and capital

World UK US

Complements

Substitute <1

Substitute >1

given the extent to which the estimatedscope for substitution varies betweendifferent sectors and levels of aggregation(not to mention between different fuelsand types of capital), broad conclusions ofthis type could be misleading. Moreover,given the difficulties with the empiricalliterature, very little confidence can beplaced in this result.

5.5 Energy productivity andecological economicsIn his 1984 paper, Brookes quotes SamSchurr’s observation that: “….it is energythat drives modern economic systemsrather than such systems creating ademand for energy.” (Brookes, 1984).This highlights an important theme inBrookes’ work: namely that energy plays

S4134 12/10/07 13:56 Page 77

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

78

a more important role in economic growththan is conventionally assumed byneoclassical economists. But precisely thesame claim is made by ecologicaleconomists such as Cleveland, et al.(1984), who attribute a large componentof the increased productivity over the pastcentury to the increasing availability ofhigh-quality energy sources. This leadsthem to express scepticism over the scopefor decoupling economic growth fromincreased energy consumption.

Ecological economists have not directlyinvestigated the rebound effect, but theirwork arguably provides suggestivesupport for the K-B postulate in much thesame way as Schurr’s research on thehistorical determinants of US productivitygrowth. Four examples of this work arebriefly described below.

First, analysis by Kaufmann (1992; 2004)and others suggests that historicalreductions in energy/GDP ratios owemuch more to structural change andimprovements in energy quality than totechnological improvements in energyefficiency (Box 5.7). By neglectingchanges in energy quality, conventionalanalysts may have come to incorrectconclusions regarding the rate anddirection of technical change and itscontribution to reduced energyconsumption. Kaufmann (1992) suggeststhat, not only does the energy/GDP ratioreflect the influence of factors other thanenergy-saving technical change, but theseother factors may be sufficient to explainthe observed trends. Hence, the observedimprovements in the thermodynamicefficiency of individual devices at themicro level do not appear to havesignificantly contributed to the observed

reduction in energy intensity at themacro-level. As with the work ofJorgenson and others on energy-usingtechnical change, this suggests that theconventional assumptions of energy-economic models may be flawed.

Second, both neoclassical and ecologicaleconomists have used moderneconometric techniques to test thedirection of causality between energyconsumption and GDP (Stern, 1993;Chontanawat, et al., 2006; Lee, 2006;Yoo, 2006; Zachariadis, 2006). If GDPgrowth is the cause of increased energyconsumption then a change in the growthrate should be followed by a change inenergy consumption and vice versa. It isargued that if causality runs from GDP toenergy consumption then energyconsumption may be reduced withoutadverse effects on economic growth, whileif causality runs the other way round areduction in energy use may negativelyaffect economic growth. While the resultsof such studies are frequentlycontradictory, most of them neglectchanges in energy quality. When energyquality is taken into account, the causalityappears to run from energy consumptionto GDP - as ecological economists suggest(Stern, 1993; 2000).

Third, historical experience provides verylittle support for the claim that increasesin income will lead to declining energyconsumption (Stern, 2004b; Richmondand Kaufmann, 2006b; 2006a). While theincome elasticity of aggregate energyconsumption may be both declining andless than one in OECD countries, there isno evidence that it is negative (or is soonto become negative). Again, neglect ofchanges in fuel mix and energy prices

S4134 12/10/07 13:56 Page 78

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

79

Kaufmann (1992) sought to quantify the factors that contributed to changes in the ratio of primaryenergy consumption (in kWh thermal) to real GDP in France, Germany, Japan and the UK during theperiod 1950-1990. The explanatory variables were the percentage share of different energy carriersin primary energy consumption; the fraction of GDP spent directly on energy by households; theproportion of the product mix that originated in energy intensive manufacturing sectors; and primaryenergy prices.

Despite the simplicity of this formulation, it was found to account for most of the variation in energyintensity for the four countries studied throughout the post-war period. Kaufmann argued thatimprovements in energy quality led to lower energy intensities by allowing more useful work to beobtained from each heat unit of energy input. The shift from coal to oil contributed greatly todeclining energy/GDP ratios prior to 1973, while the rising contribution of primary electricity (hydroand nuclear) provided a significant contribution after 1973.

Since the energy intensity of household energy purchases is an order of magnitude greater than theenergy intensity of other goods and services, falls in the former as a fraction of total expenditureshould translate into falls in the energy/GDP ratio – and vice versa. The fraction of GDP spent directlyon energy by households increased prior to 1973 and decreased thereafter and these trends werealso found to be highly significant in explaining trends in the aggregate ratio.

In addition, changes in energy prices encouraged substitution between inputs, including thesubstitution of capital for energy, while shifts towards less energy intensive manufacturing sectorsand towards the service sector reduced energy/GDP ratios. These mechanisms were found to be lessimportant than those above, but when all four factors were taken into account, they were found toprovide a more or less sufficient explanation for the observed trends in energy intensity.

By implication, Kaufmann’s results suggest little role for energy-saving technical change - defined asadvances in technology that allow the same type and quantity of output to be produced with lessenergy inputs. Kaufmann tested this implication in three different ways,52 but in each case failed tofind statistically significant evidence for energy saving technical change. Kaufmann comments that“….Technical changes has reduced the amount of energy (as measured in heat units) used to producea unit of output. But characterising that technical change as ‘energy-saving’ is misleading. Over thelast 40 years, technical change has reduced the amount of energy use to produce a unit of outputby developing new techniques for using oil, natural gas and primary electricity in place of coal.”

Kaufmann also interprets the results as illustrating the limited scope, at the level of the macro-economy, for substituting capital and labour for energy. Estimated annually, the own price elasticityof energy demand varies between -0.05 and -0.39, which is generally smaller than the elasticitiesestimated at the level of individual sectors. This arguably suggests that the indirect energyconsumption associated with labour and capital inputs constitute a significant portion of the energysaved directly through energy efficiency improvements in each of those sectors.

The results also indicate that reducing the fraction of GDP spent directly on energy by households,may be the most effective way of reducing the energy/GDP ratio. This in turn suggests that reboundeffects from energy efficiency improvements may be lower in the household sector than in producingsectors.

Box 5.7 Energy/GDP ratios and changes in energy quality

52 Namely: a) seeking evidence for serial correlation and heteroscedasticity in the error term, which could be evidence of missing variablebias; b) including a time trend to represent energy-saving technical change; and c) using dummy variables to test for changes in the interceptor slope of individual regression coefficients during different time periods - such as may follow an increase in energy prices if this inducesenergy saving technical change.

S4134 12/10/07 13:56 Page 79

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

80

may have led earlier studies to drawmisleading conclusions regarding the roleand potential of energy saving technicalchange (Kaufmann, 2004).

Finally, ecological economists havedeveloped a number of alternatives to theconventional models of economic growth(Kummel, et al., 1985; Beaudreau, 1998;Kummel, et al., 2000; Ayres and Warr,2005). A key feature of these models is adeparture from the traditional assumptionthat the productivity of each input isproportional to the share of that input inthe value of output. Instead, theproductivity of each input is estimateddirectly from a production function. Thesemodels are found to reproduce historicaltrends in economic growth extremely well,without attributing any role to technicalchange. This is in contrast to conventionaltheories of economic growth, whichattribute much of the increase in output totechnical change.53 The marginalproductivity of energy inputs is found tobe around ten times larger than its costshare, implying that improvements in theproductivity of energy inputs could have adramatic effect on economic growth andtherefore on economy-wide energyconsumption – in other words, therebound effect could be very large.

Of particular interest is the work by Ayresand Warr (2005), who combine historicaldata on the exergy content of fuel inputsand thermodynamic (second-law)conversion efficiencies (Table 5.2) to

develop a unique time series of the exergyoutput of conversion devices (termeduseful work) in the US economy over thepast century. They show that useful workinputs to the US economy have grown bya factor of 18 over the past 100 years,implying that the useful work obtainedfrom fuel resources has grown muchfaster than the consumption of fuelsthemselves, owing to substantialimprovements in thermodynamicconversion efficiencies. This approachmakes a great deal of sense, since ituseful work that is economicallyproductive, while the exergy that is lost inconversion processes is effectively wasted(Ayres and Warr, 2006).

By including useful work in theirproduction function, rather than primaryenergy, Ayres and Warr obtain anextremely good fit to US GDP trends overthe past century, thereby eliminating theneed for a multiplier for technical change.The implication is that improvements inthermodynamic conversion efficiencyprovide a quantifiable surrogate for allforms of technical change that contributeto economic growth. Far from being aminor contributor to economic growth,improvements in thermodynamicefficiency become the dominant driver –obviating the need for alternativemeasures of technological change.

While firmly outside mainstreameconomics, the ecological perspective iswell articulated and persuasive. The

53 Traditional neoclassical growth models estimate ‘technical change’ as the residual growth in output that is not explained bythe growth of inputs. Early growth models attributed as much as 70% of the growth in output to technical change, but laterstudies have shown how the proportion of growth that is attributed to technical change depends upon how the inputs aremeasured (Jorgenson and Griliches, 1967). In neoclassical growth models, technical change is typically represented by a simpletime trend. Modern theories of economic growth seek to make the source and direction of technical change endogenous.

S4134 12/10/07 13:56 Page 80

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

81

general implication is that energy is moreproductive than is suggested by its smallshare of total costs. This is precisely theargument that Schurr made and whichappears to underlie Brookes’ arguments infavour of backfire. However, the empiricalevidence in support of this perspectiveremains patchy and in some cases flawed.For example, the results of econometricinvestigations of causality relationshipsbetween energy and GDP remainambiguous and the policy implicationsthat are drawn are frequentlyoversimplified (Zachariadis, 2006). Also,the statistical form of causality that isbeing measured here (so-called ‘Grangercausality’) is not the same as causality asconventionally understood54 andconventional notions of causality may beproblematic for systems as complex asmodern economies. In a similar manner,the different variants of ‘ecological growthmodels’ rely upon an unusual and oddly

behaved production function55, provideresults that are difficult to reconcile witheach other56 and appear vulnerable tobias from a number of sources57 thatcould potentially invalidate the results. Asa result, claims that the marginalproductivity of energy is an order ofmagnitude larger than its cost share, orthat improvement in thermodynamicconversion efficiency can act as a suitableproxy for technical change, must betreated with considerable caution.

Unfortunately, the different assumptionsof conventional and ecologicalperspectives seem to have prevented anobjective comparison of their methodsand conclusions. Convincing evidence ofthe disproportionate contribution ofenergy to economic growth thereforeremains elusive. Moreover, even if thiswere to be accepted, the link from thisevidence to the K-B postulate remainsambiguous and indirect.

Year Electricity Transportation High Medium Lowgeneration temperature temperature temperature

and process heat process heat space heatdistribution (steel) (steam)

1900 3.8 3.0 7 5 0.25

1970 32.5 8.0 20 14 2

1990 33.3 13.9 25 20 3

54 For example, a met office prediction of rain can be shown to ‘Granger cause’ rain!

55 The so-called LINEX production function implies increasing marginal returns and variable marginal productivities.

56 Ayres and Warr (2005) claim that the inclusion of useful work rather than primary energy in the production function allowsthem to dispense with a separate time trend to represent technical change. But they made no comment as to why Kummel(1985; 2000) is able to reproduce economic growth without a time trend, while measuring energy inputs on the basis of thermalcontent.

Table 5.2 Trends in second-law conversion efficiencies of primary conversionprocesses in the US (average % efficiency in specified year)

Source: Ayres et al. (2003)

S4134 12/10/07 13:56 Page 81

The neoclassical assumption appears tobe that capital, labour and energy inputshave independent and additive effects oneconomic output, with any residualincrease being attributed to exogenoustechnical change. Endogenous growththeory has modified these assumptions,but still attributes a relatively minor roleto energy. In contrast, the ecologicalassumption appears to be that capital,labour and energy are interdependentinputs that have synergistic andmultiplicative effects on economic output,and that the increased availability of low-cost, high-quality energy sources providesa necessary condition for technicalchange. A bridge between the two couldpotentially be provided by Toman andJemelkova’s (2003) observation thatincreased inputs of useful work (or energyservices) may enhance the productivity ofcapital and labour:

“……when the supply of energy services isincreased, there is not just more energyto be used by each skilled worker ormachine; the productivity with whichevery unit of energy is used also rises. Ifall inputs to final production are increasedin some proportion, final output wouldgrow in greater proportion because of theeffect on non-energy inputs.” (Toman andJemelkova, 2003)

Focusing in particular on households indeveloping countries, Toman andJemolkova (2003) propose a number ofways in which the increased availability ofuseful work could improve capital andlabour productivity and hencedisproportionately affect economic output.For example, cheaper and better lightingcould allow greater flexibility in timeallocation throughout the day and eveningand enhance the productivity of education

efforts. The increased availability ofelectricity could promote access to safedrinking water (e.g. in deeper wells),allow the refrigeration of food andmedicine and thereby improve both thehealth of workers and their economicproductivity. Similarly, the increasedavailability of low-cost transport fuelscould interact with investment in transportinfrastructure to increase the geographicsize, scale and efficiency of markets.Schurr’s (1983; 1984; 1985) account ofthe impact of electricity (and especiallyelectric motors) on the organisation andproductivity of US manufacturing providesan analogous example for producers indeveloped countries.

It is an empirical question as to whethersuch benefits apply in practice and towhat extent. Ecological economistsappear to claim that such a situation is thenorm, with the result that the increasedavailability of high-quality energy hasbeen a primary driver of economicactivity. But if the increased availability ofhigh-quality energy inputs has adisproportionate impact on productivityand economic growth, then improvementsin thermodynamic efficiency may do thesame, because both increase the usefulwork available from conversion devices. Ifit is useful work rather than raw energy(or exergy) inputs that drives economicactivity, then improvements in second-lawconversion efficiency could potentiallymitigate the economic impact of futureshortages of high-quality forms of energy– notably oil.

However, this argument uses athermodynamic measure of energy qualityand therefore neglects the other factorsthat determine the economic productivity

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

82

S4134 12/10/07 13:56 Page 82

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

83

of energy carriers. Also, improvements inconversion efficiencies are necessarilyassociated with embodied energy and areultimately constrained by thermodynamiclimitations; hence, their ability tocompensate for future supply shortagesmay be limited. Moreover, if theseimprovements have a disproportionateeffect on economic output, they may alsobe associated with large secondaryeffects.

5.6 Implications One interpretation of the K-B postulate isthat all economically justified energyefficiency improvements will increaseenergy consumption above where it wouldbe without those improvements. This is acounterintuitive claim for many peopleand therefore requires strong supportingevidence if it is to gain widespreadacceptance. The main conclusion from thereview is that such evidence does notexist. The theoretical and empiricalevidence cited in favour of the postulate issuggestive rather than definitive, onlyindirectly relevant to the rebound effectand flawed in a number of respects.Nevertheless, the arguments andevidence deserve more serious attentionthan they have received to date. Much ofthe evidence points to economy-widerebound effects being larger than isconventionally assumed and to energyplaying a more important role in economicgrowth than is conventionally assumed.

The possibility of large economy-widerebound effects has been dismissed by anumber of leading energy analysts(Howarth, 1997; Lovins, 1998; Laitner,

2000; Schipper and Grubb, 2000). But itbecomes more plausible if it is acceptedthat energy efficiency improvements arefrequently associated with improvementsin the productivity of other inputs. If thisis the case, then rebound effects need notnecessarily be small just because theshare of energy in total costs is small.Future research should thereforeinvestigate the extent to whichimprovements in energy efficiency(however defined and measured) areassociated with broader improvements ineconomic productivity, and thecircumstances under which economy-widerebound effects are more or less likely tobe large. For example, we may speculatethat rebound effects should be larger forenergy efficiency improvementsassociated with:

• energy intensive production sectorscompared to non-energy intensivesectors;

• energy supply industries compared toenergy users;

• core process technologies comparedto non-core technologies;

• technologies in the early stages ofdiffusion compared to those in thelater stages; and

• technologies that improve capital andlabour productivity, compared tothose that do not.

Rebound effects may be particularly largefor the energy efficiency improvementsassociated with ‘general-purposetechnologies’, such as steam engines,railroads, automobiles and computers.General-purpose technologies (GPTs)

S4134 12/10/07 13:56 Page 83

have a wide scope for improvement andelaboration, are applicable across a broadrange of uses, have potential for use in awide variety of products and processesand have strong complementarities withexisting or potential new technologies(Lipsey, et al., 2005). Steam enginesprovide a paradigmatic illustration of aGPT in the 19th-century, while electricmotors provides a comparable illustrationfor the early 20th century. The former wasused by Jevons to support the case forbackfire, while the latter was used byBrookes.

The key to unpacking the K-B postulatemay therefore be to distinguish theenergy efficiency improvementsassociated with GPTs from other forms ofenergy efficiency improvement. The K-Bpostulate seems more likely to hold forthe former, particularly when these areused by producers and when the energyefficiency improvements occur at an earlystage of development and diffusion of thetechnology. The opportunities offered bythese technologies have such long termand significant effects on innovation,productivity and economic growth thateconomy-wide energy consumption isincreased. In contrast, the K-B postulateseems less likely to hold for dedicatedenergy efficiency technologies such asimproved thermal insulation, particularlywhen these are used by consumers orwhen they play a subsidiary role ineconomic production. These technologieshave smaller effects on productivity andeconomic growth, with the result thateconomy-wide energy consumption maybe reduced.

The implication is that energy policyshould focus on encouraging dedicated

energy efficient technologies, rather thanimproving the energy efficiency of GPTs.However, these categories are poorlydefined and the boundaries between themare blurred. Moreover, even if GPTs canmeaningfully be distinguished from otherforms of technology, continued economicgrowth is likely to depend upon thediffusion of new types of GPT that mayincrease aggregate energy consumption.Hence, while it may be unlikely that allenergy efficiency improvements will leadto backfire, we still have much to learnabout the factors that make backfire moreor less likely.

5.7 Summary• The case for the K-B postulate is not

based upon empirical estimates ofrebound effects, but instead reliesupon stylised theoretical argumentsand empirical evidence from a rangeof sources that is both suggestive andindirect. Disputes over the postulaterest in large part on competingtheoretical assumptions.

• Historical experience demonstratesthat substantial improvements invarious measures of energy efficiencyhave occurred alongside continuingincreases in economic output, totalfactor productivity and overall energyconsumption. However, the causallinks between these trends remainsunclear.

• Brookes has developed a range ofarguments in support of the K-Bpostulate and cited a number ofsuggestive sources of empiricalevidence. However, his theoretical

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

84

S4134 12/10/07 13:56 Page 84

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

85

arguments have a number ofweaknesses and more recentempirical evidence does not alwayssupport his case.

• A key theme in Brookes’ work is thatimprovements in energy productivityare generally associated withproportionally greater improvementsin total factor productivity. Evidencefor this can be found at the level ofnational economies, as well asindividual sectors and technologies. Ifenergy efficient technologies boostthe productivity of other inputs andthereby save costs on more thanenergy alone, the argument thatrebound effects must be smallbecause the share of energy in totalcosts is small is undermined. But thisleaves open the question of whetherthis is necessarily the case, orwhether it is contingent uponparticular circumstances.

• Saunders has shown how backfire isthe predicted outcome of neoclassicalproduction functions that are usedwidely in theoretical and empiricalresearch. If such functions areconsidered to provide a reasonablerepresentation of real-worldbehaviour, Saunders’ work suggeststhat ‘pure’ energy-efficiencyimprovements are likely to lead tobackfire. Alternatively, if reboundeffects vary widely in magnitudebetween different sectors, suchfunctions cannot be used to representthem. In either case, the implicationsare far-reaching.

• Rebound effects depend in part uponthe scope for substitution betweenenergy and other inputs, but thenature of this relationship is morecomplex than commonly assumed.The extensive empirical literature inthis area is both confused andinconclusive and provides aninsufficient basis for the assumedparameter values within energy-economic models. To the extent thatgeneral conclusions can be drawn, it isthat capital and energy appear to becomplements or only weaksubstitutes. This suggests thepossibility of a strong link betweenenergy consumption and economicoutput and high costs associated withreducing energy consumption.

• Underlying Brookes’ work is a claimabout the contribution of energy toeconomic growth. Most economistsassume that the increased availabilityof energy inputs has only made asmall contribution to economicgrowth, owing to the small share ofenergy in total costs. In contrast,ecological economists argue that theincreased availability and quality ofenergy inputs has been the primarydriver of economic growth. But energyis only economically productivebecause it provides useful work.Hence, if increases in energy inputscontribute disproportionately toeconomic growth, then improvementsin thermodynamic efficiency should dothe same, since both provide moreuseful work. The dispute over the K-Bpostulate can therefore be linked tothe much broader question of the

S4134 12/10/07 13:56 Page 85

contribution of energy to economicgrowth.

• Ecological economists cite a range ofevidence in support of their claims,but this remains patchy and in somecases flawed. Of particular interestare the alternatives to standardmodels of economic growth which canbe used in support of claims thatenergy efficiency improvements areassociated with substantialimprovements in total factorproductivity and that rebound effectsare large. However, since thesemodels have a number of theoreticaland empirical weaknesses, convincingevidence of the disproportionatecontribution of energy remain elusive.

• The debate over the K-B postulatewould benefit from further distinctionsbetween different types of energy

efficiency improvement. In particular,the K-B postulate seems more likelyto hold for energy efficiencyimprovements associated with theearly stage of diffusion of ‘general-purpose technologies’, such as electricmotors in the early 20th century. Itmay be less likely to hold for the laterstages of diffusion of thesetechnologies, or for ‘dedicated’ energyefficiency technologies such asimproved thermal insulation.However, these categories are poorlydefined and the boundaries betweenthem are blurred.

• Overall, while it is unlikely that allenergy efficiency improvements willlead to backfire, we still have much tolearn about the factors that makebackfire more or less likely to occur.

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

86

S4134 12/10/07 13:56 Page 86

6.1 ConclusionsThe main conclusions from thisassessment are as follows:

1. Rebound effects are significant,but they need not make energyefficiency policies ineffective inreducing energy demand

• The available evidence for all types ofrebound effects is limited andinconclusive. While the evidence isbetter for direct effects than forindirect effects, it remains mainlyfocused on a small number ofconsumer energy services withinOECD countries. Both direct andindirect rebound effects appear tovary widely between differenttechnologies, sectors and incomegroups and in most cases cannot bequantified with much confidence.

• The evidence does not suggest thatimprovements in energy efficiencyroutinely lead to economy-wideincreases in energy consumption, assome commentators have suggested.At the same time the evidence doesnot suggest that economy-widerebound effects are small (e.g.<10%) as many analysts andpolicymakers assume. Reboundeffects therefore need to be takenseriously in policy appraisal.

2. For most consumer energyservices in OECD countries, directrebound effects are unlikely toexceed 30%

• Evidence for the direct rebound effectfor personal automotive transport,household heating and household

cooling within OECD countries isrelatively robust. Evidence for directrebound effects for other consumerenergy services is much weaker, as isthat for energy efficiencyimprovements by producers.

• For household heating, householdcooling and personal automotivetransport in OECD countries, thedirect rebound effect is likely to beless than 30% and may be closer to10% for transport. Moreover, directrebound effects for these energyservices are expected to decline in thefuture as demand saturates andincome increases. This means thatimprovements in energy efficiencyshould achieve 70% or more of theexpected reduction in energyconsumption for those services -although the existence of indirecteffects means that the economy-widereduction in energy consumption willbe less. Direct rebound effects shouldbe smaller for other consumer energyservices where energy forms arelatively small proportion of totalcosts and therefore has little influenceon operating decisions.

• These conclusions are subject to anumber of important qualifications,including the dependence of directrebound effects on household income,the neglect of ‘marginal consumers’and the relatively limited time periodsover which these effects have beenstudied

• There are very few studies of directrebound effects from energy efficiencyimprovements in developingcountries. However, both theoretical

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

87

6. Conclusions, research needs and policy implications

S4134 12/10/07 13:56 Page 87

considerations and the limitedevidence that is available suggest thatdirect rebound effects in thesecontexts could be larger than in OECDcountries and could in some casesexceed unity. This is especially thecase for the 1.6 billion householdswho currently lack access to electricityand the 2.5 billion who rely uponbiomass for cooking.

3. There are relatively fewquantitative estimates of indirectand economy-wide reboundeffects, but several studiessuggest that economy-wideeffects may exceed 50%

• Quantitative estimates of indirect andeconomy-wide rebound effects arerare. While a number ofmethodological approaches can beused to estimate these effects, thelimited number of studies availableprovides an insufficient basis to drawany general conclusions. The mostimportant insight from these studiesis that the magnitude of the effectdepends very much upon the sectorwith the energy efficiencyimprovement takes place and issensitive to a number of variables.

• A handful of CGE modeling studiesestimate economy-wide reboundeffects to be 37% or more, with halfof the studies predicting backfire.These effects derive from ‘pure’energy efficiency improvements byproducers (not consumers) andtherefore do not rely uponsimultaneous improvements in theproductivity of other inputs. However,the small number of studies available,

the diversity of approaches used andthe variety of methodologicalweaknesses associated with the CGEapproach all suggest the need forcaution when interpreting theseresults.

• In principle, more robust estimates ofthe economy-wide rebound effect maybe obtained from macro-econometricmodels of national economies. Barkerand Foxon (2006) use this approachto estimate an economy-wide reboundeffect of 26% from current UK energyefficiency policies. However, there area number of reasons why this studycould have underestimated economy-wide effects.

4. The evidence and argumentsused in support of the Khazzoom-Brookes postulate are insufficientto demonstrate its validity, butnevertheless pose an importantchallenge to conventionalwisdom

• The theoretical arguments for the K-Bpostulate rely upon a conceptualframework that is stylised andrestrictive, while the empiricalevidence cited in its favour is indirectand suggestive. Since a number offlaws have been found with both, theK-B ‘hypothesis’ cannot be consideredto have been adequately verified.Nevertheless, the arguments andevidence used to defend the K-Bpostulate deserve more seriousattention than they have received todate.

• It is conventionally assumed thatthere is considerable scope for

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

88

S4134 12/10/07 13:56 Page 88

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

89

substituting capital and other inputsfor energy consumption whilemaintaining the same level ofeconomic output. It is alsoconventionally assumed that technicalchange has historically improved theenergy efficiency of individual sectorsand thereby contributed to theobserved decoupling of energyconsumption from economic growth.However, the evidence reviewed inthis report suggests that there is morelimited scope for substituting otherinputs for energy and that muchtechnical change has acted to increaseenergy intensity. Also, once differentfuels are weighted by their relative‘quality’ or economic productivity,there is less evidence that the growthin economic output has beendecoupled from the growth in energyconsumption. Overall, this evidencepoints to economy-wide reboundeffects relatively large and to energyplaying a more important role ineconomic growth than isconventionally assumed.

• The possibility of large economy-widerebound effects becomes moreplausible if it is accepted that energyefficiency improvements arefrequently associated withproportionately greater improvementsin total factor productivity. If this isthe case, then rebound effects neednot necessarily be small just becausethe share of energy in total costs issmall. But energy efficiencyimprovements may not necessarily beassociated with such improvements.Instead, the link between the twoseems more likely to be contingent

upon particular technologies andcircumstances.

• The debate over the K-B postulatewould benefit from more carefuldistinctions between different types ofenergy efficiency improvement. Forexample, the K-B postulate seemsmore likely to hold for energyefficiency improvements associatedwith ‘general-purpose technologies’(GPTs), particularly when these areused by producers and when theimprovements occur at an early stageof development and diffusion. Steamengines provide a paradigmaticillustration of a GPT in the 19th-century, while electric motors providea comparable illustration for the early20th century. The opportunitiesoffered by these technologies havesuch long term and significant effectson innovation, productivity andeconomic growth that economy-wideenergy consumption is increased. Incontrast, the K-B postulate seems lesslikely to hold for dedicated energyefficiency technologies such asthermal insulation, particularly whenthese are used by consumers. Thesetechnologies have smaller effects onproductivity and economic growth,with the result that economy-wideenergy consumption may be reduced.

6.2 Research needsGiven the potential importance of reboundeffects, the evidence base is remarkablyweak. While this is partly a consequenceof the inherent difficulty of measuring orestimating such effects, there is

S4134 12/10/07 13:56 Page 89

considerable scope for improvingknowledge in a number of areas. Thefollowing highlights some priorities forfuture research.

1. Research on direct reboundeffects needs to improve inrigour and expand in scope

• Estimates of the direct rebound effectare contingent upon good data setsand would benefit from more robustmethodologies. This is particularlycase for quasi-experimental studies ofhousehold heating, where currentevaluation practice is poor.

• Econometric studies need to addressthe potential sources of bias indicatedin Section 3. There is scope for botheconometric and quasi-experimentalstudies of a greater range ofconsumer energy services, providedthat individual appliances can bemonitored. However, the policy issuehere is not so much changes in short-term utilisation patterns, but changesin the number and capacity ofconversion devices over the longerterm.

• Estimates of the direct rebound effectfor personal automotive transportwould benefit from more appropriatedefinitions of useful work. A studyemploying tonne kilometres as thedependent variable appears feasibleand could potentially capture theeffect of increasing car sizes. Analysisis also needed of other modes oftransport, including freight.

• There is scope for more empiricalwork on the ‘rebound effect withrespect to time’, especially in the areaof transportation. More work is also

required on the dependence of directrebound effects on income.

• The geographical bias of the evidencebase also needs to be addressed. Inparticular, the evidence for reboundeffects in developing countries is veryweak.

2. Quantitative estimates of indirectand economy-wide reboundeffects are feasible and should bepursued

• A combination of input-outputanalysis and life-cycle analysis can beused to estimate the embodied energyassociated with various types ofenergy efficiency improvement. Theseestimates need to be developed moresystematically and the resultsincorporated within technology andpolicy appraisals.

• Embodied energy analysis can also becombined with econometric models ofconsumer or producer behaviour toestimate the secondary effects fromenergy efficiency improvements byhouseholds. This approach is limitedto rather aggregate categories ofenergy service and has a number ofmethodological weaknesses.Nevertheless, there is considerablescope for further research.

• Given the widespread use of CGEmodels in energy research, the lack ofapplication to rebound effects issurprising. There is much scope forexpanding the evidence base in termsof the countries and sectors studiedand the different types of energyefficiency improvement that aremodelled. However, the empiricalbasis for CGE models needs to be

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

90

S4134 12/10/07 13:56 Page 90

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

91

improved and there is a need for moresystematic and informed sensitivityanalysis to highlight the importance ofspecific assumptions. This would helpestablish how robust the results are,and with what confidence they can beexpressed.

• More robust estimates of economy-wide rebound effects may potentiallybe obtained from macro-econometricmodels. There is scope for furtherapplication of this methodology, inwhich some of the weaknessesidentified here could be addressed.

3. Our understanding of thecontribution of energy toeconomic growth needs to begreatly improved

• The linkage between economicmeasures of energy efficiency at themacro-level and physical/thermodynamic measures of energyefficiency at the micro level is poorlyunderstood. While decompositionanalysis is routinely used to identifythe relative contribution of structuralchange and changes in variousmeasures of energy efficiency, therelative importance of technicalchange, substitution between energyand other inputs, and changes in fuelmix remains unclear. Since standardassumptions about ‘autonomousenergy efficiency improvements’(AEEI) appear inconsistent with theavailable evidence, the projections ofmany energy-economic models maybe misleading.

• Since thirty years of research hasfailed to reach a consensus on the

issue of substitutability betweenenergy and capital, the underlyingtheoretical assumptions and/ormethodological approach may beflawed. Future work should ensurethat restrictions such as neutraltechnical change are tested for ratherthan assumed and only accepted onempirical grounds. Future studiesshould also use a flexible form for theproduction function, allow for non-neutral technical change, test theassumption of ‘separability’ betweendifferent inputs and pay closerattention to changes in product mix. Afull meta-analysis of existing studieswould also be beneficial, to furtherclarify the reasons for the differingresults.

• Future research should investigatewhether, how and to what extentdifferent types of energy efficiencyimprovement at different levels ofaggregation are associated withimprovements in the productivity ofother inputs and with improvementsin total factor productivity.

• ‘Ecological’ models of economicgrowth challenge conventionalassumptions and offer a promisingroute for further research. At present,this approach is largely ignored bymainstream economists, whogenerally pay insufficient attention tothe contribution of energy toeconomic growth. While Saunders hasused neoclassical growth theory toexplore the rebound effect, the issuehas not been addressed bycontemporary research onendogenous growth theory andinduced technical change. While

S4134 12/10/07 13:57 Page 91

communication is inhibited in part bycompeting ‘world-views’, there shouldbe scope for mutual learning andimproved testing.

6.3 Policy implicationsThe general conclusion of this assessmentis that rebound effects need be takenmore seriously by analysts andpolicymakers than has hitherto been thecase. While the effectiveness of energyefficiency policies has not beeninvestigated, some general policyimplications can nevertheless be drawn:

1. The potential contribution ofenergy efficiency policies needsto be reappraised.

• Energy efficiency may be encouragedthrough policies that raise energyprices, such as carbon taxes, orthrough non-price policies such asbuilding regulations. Both shouldcontinue to play an important role inenergy and climate policy. However,many official and independentappraisals of such policies haveoverstated the potential contributionof non-price policies to reducingenergy consumption and carbonemissions.

• It would be wrong to assume that, inthe absence of evidence, reboundeffects are so small that they can bedisregarded. Under somecircumstances (e.g. energy efficienttechnologies that significantlyimprove the productivity of energyintensive industries) economy-wide

rebound effects may exceed 50% andcould potentially increase energyconsumption in the long-term. Inother circumstances (e.g. energyefficiency improvements in consumerelectronic goods) economy-widerebound effects are likely to besmaller. But in no circumstances arethey likely to be zero.

• Taking rebound effects into accountwill reduce the apparent effectivenessof energy efficiency policies. However,many energy efficiency opportunitiesare highly cost-effective and willremain so even when rebound effectsare allowed for. Provided market andorganisational failures can beovercome, the encouragement ofthese opportunities should increasereal income and contribute toeconomic growth. They may not,however, reduce energy consumptionand carbon emissions by as much aspreviously assumed.

2. Rebound effects should be takeninto account when developingand targeting energy efficiencypolicy

• Rebound effects vary widely betweendifferent technologies, sectors andincome groups. While thesedifferences cannot be quantified withmuch confidence, there should bescope for including estimated effectswithin policy appraisals and usingthese estimates to target policiesmore effectively. Where reboundeffects are expected to be large, theremay be a greater need for policiesthat increase energy prices.

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

92

S4134 12/10/07 13:57 Page 92

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

93

• ‘Win-win’ opportunities that reducecapital and labour costs as well asenergy costs may be associated withlarge rebound effects. Hence, theimplications of encouraging theseopportunities need to be clearlyunderstood and quantified. It maymake more sense to focus policy on‘dedicated’ energy efficienttechnologies, leaving the realisation ofwider benefits to the market

3. Rebound effects may bemitigated through carbon/energypricing – whether implementedthrough taxation or an emissionstrading scheme

• Carbon/energy pricing can reducedirect and indirect rebound effects byensuring that the cost of energyservices remains relatively constant

while energy efficiency improves.Carbon/energy pricing needs toincrease over time at a rate sufficientto accommodate both income growthand rebound effects, simply toprevent carbon emissions fromincreasing. It needs to increase morerapidly if emissions are to be reduced.

• Carbon/energy pricing may beinsufficient on its own, since it will notovercome the numerous barriers tothe innovation and diffusion of lowcarbon technologies and could haveadverse impacts on incomedistribution and competitiveness.Similarly, policies to address marketbarriers may be insufficient, sincerebound effects could offset much ofthe energy savings. A policy mix isrequired.

S4134 12/10/07 13:57 Page 93

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

94

S4134 12/10/07 13:57 Page 94

Adams, F. G. and P. Miovic, (1968), 'On relative fuel efficiency and the output elasticity of energy

consumption in Western Europe', The Journal of Industrial Economics, 17(1), 41-56.

Alcorn, J. A. and G. Baird, (1996), Embodied energy analysis of New Zealand building materials-

methods and results, Proceedings of the Embodied Energy: The Current State of Play Seminar,

Deakin University, Geelong, Australia, 28-29 November

Alcott, B., (2005), 'Jevons' paradox', Ecological Economics, 54(1), 9-21.

Alfredsson, E. C., (2004), ''Green' consumption - no solution for climate change', Energy, 29, 513-

24.

Allan, G., N. Hanley, P. G. McGregor, J. Kim Swales, and K. Turner, (2006), 'The macroeconomic

rebound effect and the UK economy', Final report to the Department Of Environment Food and Rural

Affairs, Department Economics, University of Strathclyde, Strathclyde.

Ang, B. W., (1999), 'Decomposition methodology in energy demand and environmental analysis', in

Handbook of Environmental and Resource Economics, J.C.J.M van der Bergh ed, Edward Elgar,

Cheltenham, UK.

Ang, B. W., (2006), 'Monitoring changes in economy-wide energy efficiency: From energy-GDP ratio

to composite efficiency index', Energy Policy, 34(5), 574-82.

Atack, J., (1979), 'Fact in fiction? The relative cost of steam and water power: a simulation

approach', Explorations in Economic History, 16, 409-37.

Ayres, R. U., (2002), 'Resources, scarcity, technology and growth', INSEAD Working Paper No.

2002/118/EPS/CMER, Centre for the Management of Environmental Resources, INSEAD,

Fontainebleau.

Ayres, R. U., L. W. Ayres, and B. Warr, (2003), 'Exergy, power and work in the US economy, 1900-

1998', Energy, 28(3), 219-73.

Ayres, R. U. and B. Warr, (2002), 'Two paradigms of production and growth', CMER Working Paper,

INSEAD, Fontainbleau, France.

Ayres, R. U. and B. Warr, (2005), 'Accounting for growth: the role of physical work', Structural

Change and Economic Dynamics, 16(2), 181-209.

Ayres, R. U. and B. Warr, (2006), 'REXS: a forecasting model for assessing the impact of natural

resource consumption and technological change from economic growth', Structural Change in

Economic Dynamics, 17(3), 329-78.

Baker, P., Blundell, R., and J. Mickelwright, (1989), 'Modelling household energy expenditures using

micro-data', Economic Journal, 99(720-738).

Baker, P. and R. Blundell, (1991), 'The microeconometric approach to modelling energy demand:

some results for UK households', Oxford Review of Economic Policy, 7(2), 54-76.

Barker, T., (2005), 'The transition to sustainability: a comparison of general equilibrium and space-

time economics approaches', Working Paper Number 62, Tyndall Centre for Climate Change

Research.

Barker, T., P. Ekins, and T. Foxon, (2007), 'Macroeconomic effects of efficiency policies for energy-

intensive industries: the case of the UK Climate Change Agreements, 2000-2010', Energy

Economics, 29(5), 760-78.

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

95

References

S4134 12/10/07 13:57 Page 95

Barker, T., P. Ekins, and N. Johnstone, (1995), Global warming and energy demand, Routledge,

London.

Barker, T. and T. Foxon, (2006), 'The macroeconomic rebound effect and the UK economy', Report

to the Department of the Environment, food and Rural Affairs, 4CMR, Cambridge.

Barro, R. J. and X. Sala-I-Martin, (1995), Economic Growth, McGraw-Hill, New York.

Beaudreau, B. C., (1998), Energy and Organisation: group and distribution re-examined,

Greenwood Press, Westport, CT.

Becker, G. S., (1965), 'A theory of the allocation of time', The Economic Journal, LXXV(299), 493-

517.

Bergh, J. C. J. M. v. d. and K. J. Button, (1999), 'Meta-analysis, economic valuation and

environmental economics', in Handbook of Environmental Resources, J.C.J.M. van den Bergh ed,

Edward Elgar, Cheltenham.

Berndt, E. R., (1978), 'Aggregate energy, efficiency and productivity measurement', Annual Review

of Energy, 3, 225-73.

Berndt, E. R., (1990), 'Energy use, technical progress and productivity growth: a survey of economic

issues', The Journal of Productivity Analysis, 2, 67-83.

Berndt, E. R. and D. O. Wood, (1979), 'Engineering and econometric interpretations of energy-

capital complementarity', American Economic Review, June, 259-68.

Berry, L., (1983), 'Residential conservation programme impacts: methods of reducing self-selection

bias', Evaluation Review, 7(6), 753-75.

Besen, S. M. and L. L. Johnson, (1982), 'Comment on "Economic implications of mandated efficiency

standards for household appliances"', Energy Journal, 3(1), 110-16.

Bhattacharyya, S. C., (1996), 'Applied general equilibrium models for energy studies: a survey',

Energy Economics, 18, 145-64.

Binswanger, M., (2001), 'Technological progress and sustainable development: what about the

rebound effect?' Ecological Economics, 36(1), 119-32.

Birol, F. and J. H. Keppler, (2000), 'Prices, technology development and the rebound effect', Energy

Policy, 28(6-7), 457-69.

Boardman, B. and G. Milne, (2000), 'Making cold homes warmer: the effect of energy efficiency

improvements in low-income homes', Energy Policy,, 218(6-7), 411-24.

Boyd, G. A. and J. X. Pang, (2000), 'Estimating the linkage between energy efficiency and

productivity', Energy Policy, 28, 289-96.

Brännlund, R., T. Ghalwash, and J. Nordstrom, (2007), 'Increased energy efficiency and the rebound

effect: effects on consumption and emissions', Energy Economics, 29(1), 1-17.

Brookes, L., (2004), 'Energy efficiency fallacies—a postscript', Energy Policy, 32(8), 945-47.

Brookes, L. G., (1972), 'More on the output elasticity of energy consumption', The Journal of

Industrial Economics, 21(1), 83-92.

Brookes, L. G., (1978), 'Energy policy, the energy price fallacy and the role of nuclear energy in the

UK', Energy Policy, 6(2), 94-106.

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

96

S4134 12/10/07 13:57 Page 96

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

97

Brookes, L. G., (1984), 'Long-term equilibrium effects of constraints in energy supply', in The

Economics of Nuclear Energy, Leonard Brookes and H. Motamen eds, Chapman and Hall, London.

Brookes, L. G., (1990), 'The greenhouse effect: the fallacies in the energy efficiency solution.'

Energy Policy, 18(2), 199-201.

Brookes, L. G., (1992), 'Energy efficiency and economic fallacies - a reply', Energy Policy, 20(5),

390-92.

Brookes, L. G., (1993), 'Energy efficiency fallacies - the debate concluded', Energy Policy, 21(4),

346-47.

Brookes, L. G., (2000), 'Energy efficiency fallacies revisited', Energy Policy, 28(6-7), 355-66.

Bullard, C. W., P. S. Penner, and D. A. Pilati, (1978), 'Net energy analysis: handbook for combining

process and input–output analysis', Resources and Energy 1 (1978) (3), pp. 267–313, 1(3),

267–313.

Cambridge Econometrics, (2005), 'Modelling the initial effects of the Climate Change Levy', HM

Customs and Excise, London.

Casals, X. G., (2006), 'Analysis of building energy regulation and certification in Europe: their role,

limitations and differences', Energy and Buildings, 38, 381-92.

Chalkley, A. M., E. Billett, and D. Harrison, (2001), 'An investigation of the possible extent of the

respending rebound effect in the sphere of consumer products', The Journal of Sustainable Product

Design, 1(163-170).

Chapman, P., (1974), 'Energy costs: a review of methods', Energy Policy, 2(2), 91-103.

Chontanawat, J., L. Hunt, C.,, and R. Pierse, (2006), 'Causality between energy consumption and

GDP: evidence from 30 OECD and 78 non-OECD countries', Working Paper, Department of

Economics, University of Surrey, Guildford.

Cleveland, C. J., (1992), 'Energy quality and energy surplus in the extraction of fossil fuels in the

U.S.' Ecological Economics, 6(2), 139-62.

Cleveland, C. J., R. Costanza, C. A. S. Hall, and R. K. Kaufmann, (1984), 'Energy and the US

economy: a biophysical perspective', Science, 225, 890-97.

Cleveland, C. J., R. K. Kaufmann, and D. I. Stern, (2000), 'Aggregation and the role of energy in the

economy', Ecological Economics, 32, 301-17.

Conrad, K., (1999), 'Computable general equilibrium models for environmental economics and policy

analysis', in Handbook of Environmental and Resource Economics, J.C.J.M van dn Bergh ed, Edward

Elgar, Cheltenham, UK.

Costanza, R., (1980), 'Embodied energy and economic valuation', Science, 210, 1219-24.

Crafys, N., (2003), 'Steam as a general-purpose technology: a growth accounting perspective',

Working Paper nO. 75/03, Department of Economic History, London School of Economics.

Dahl, C., (1993), 'A survey of energy demand elasticities in support of the development of the

NEMS', Prepared for US Department of Energy, Contract No. De-AP01-93EI23499, Department of

Mineral Economics, Colorado School of Mines, Colorado.

Dahl, C., (1994), ' Demand for transportation fuels: a survey of demand elasticities and their

components', Journal of Energy Literature, 1(2), 3.

S4134 12/10/07 13:57 Page 97

Dahl, C. and T. Sterner, (1991), 'Analyzing Gasoline Demand Elasticities - a Survey', Energy

Economics, 13(3), 203-10.

Dargay, J. M., (1992), Are price & income elasticities of demand constant? The UK experience,

Oxford Institute for Energy Studies, Oxford, U.K.

Dargay, J. M. and D. Gately, (1994), 'Oil demand in the industrialised countries', Energy Journal,

15(Special Issue), 39-67.

Dargay, J. M. and D. Gately, (1995), 'The imperfect price irreversibility of non-transportation of all

demand in the OECD', Energy Economics, 17(1), 59-71.

Davis, L. W., (2007), 'Durable goods and residential demand for energy and water: evidence from

a field trial', Working Paper, Department of Economics, University of Michigan.

Deaton, A. and J. Mulbauer, (1980), 'An almost ideal demand system', The American Economic

Review, 70, 312-26.

DEFRA, (2007), 'Consultation document: energy, cost and carbon savings for the draft EEC 2008 -

11 illustrative mix', Department of Environment, Food and Rural Affairs, London.

Denison, E. F., (1985), Trends in American Economic Growth 1929-1982, The Bookings Institute,

Washington, DC.

Department of Transport, (2006), 'National Travel Survey', Department of Transport, London.

Diewert, W. E. and T. J. Wales, (1987), 'Flexible functional forms and global curvature conditions',

Econometrica, 55(1), 43-68.

Douthitt, R. A., (1986), 'The demand for residential space and water heating fuel by energy

conserving households', The Journal of Consumer Affairs, 20(2), 231-48.

Dubin, J. A., A. K. Miedema, and R. V. Chandran, (1986), 'Price Effects of Energy-Efficient

Technologies - a Study of Residential Demand for Heating and Cooling', Rand Journal of Economics,

17(3), 310-25.

Einhorn, M., (1982), 'Economic implications of mandated efficiency standards for household

appliances: an extension.' Energy Journal, 3(1), 103-09.

EMF 4 Working Group, (1981), 'Aggregate elasticity of energy demand', Energy Journal, 2(2), 37-

75.

Engle, R. F. and C. W. J. Granger, (1987), 'Cointegration and error correction: representation,

estimation and testing', Econometricaa, 55(2), 251-76.

Espey, J. A. and M. Espey, (2004), 'Turning on the lights: a meta-analysis of residential electricity

demand elasticities', Journal of Agricultural and Applied Economics, 36(1), 65-81.

Espey, M., (1998), 'Gasoline demand revisited: an international meta-analysis of elasticities', Energy

Economics, 20, 273-95.

Fanger, O., (1970), Thermal Comfort: Analysis and Applications in Environmental Engineering,

Danish Technical Press, Copenhagen.

Feist, W., (1996), Life-cycle energy balances compared: low energy house, passive house, self-

sufficient house, Proceedings of the International Symposium of CIB W67, Vienna, Austria,

Ferrer-i-Carbonell, A., A. C. Muskens, and M. J. van Leeuwen, (2002), 'Behavioural responses to

energy-related taxes: a survey', International Journal of Global Energy Issues, 18(2/3/4), 202-17.

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

98

S4134 12/10/07 13:57 Page 98

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

99

Fisher, F. M., (1993), Aggregation: aggregate production functions and related topics, MIT Press,

Cambridge, MA.

Fouquet, R. and P. Pearson, (2006), 'Seven centuries of energy services: the price and use of light

in the United Kingdom (1300-1700)', The Energy Journal, 27(1), 139-77.

Frey, C. J. and D. G. Labay, (1988), 'Examination of energy take-back', Energy Systems and Policy,

12, 205-17.

Friedman, D., (1987), 'Cold houses in warm climates and vice versa: a paradox of rational heating',

Journal of Political Economy, 95, 1089-97.

Frondel, M., (2004), 'Empirical assessment of energy price policies: the case for cross price

elasticities', Energy Policy, 32, 989-1000.

Frondel, M. and C. M. Schmidt, (2004), 'Facing the truth about separability: nothing works without

energy', Ecological Economics, 51, 217-23.

Frondel, M. and C. M. Schmidt, (2005), 'Evaluating Environmental Programs: The Perspective of

Modern Evaluation Research', Ecological Economics, 55(4), 515-26.

Gately, D., (1992), 'Imperfect price reversed ability of US gasoline demand: asymmetric responses

to price increases and declines', Energy Journal, 13(4), 179-207.

Gately, D., (1993), 'The imperfect price reversibility of world oil demand', Energy Journal, 14(4),

163-82.

Geller, H., P. Harrington, A. H. Rosenfeld, S. Tanishima, and F. Unander, (2006), 'Policies for

increasing energy efficiency: thirty years of experience in OECD countries', Energy Policy, 34, 556-

73.

Goldberg, P. K., (1996), 'The effect of the corporate average fuel effciency standards.' Working Paper

No: 5673, National Bureau of Economic Research, Cambridge,MA.

Graham, D. J. and S. Glaister, (2002), 'The demand for automobile fuel: a survey of elasticities',

Journal of Transpor Economics and Policy, 36(1), 1-26.

Greene, D. L., (1992), 'Vehicle use and fuel economy: how big is the "rebound" effect?' Energy

Journal, 13(1), 117-43.

Greene, D. L., J. R. Kahn, and R. Gibson, (1999a), 'An econometric analysis of the elasticity of

vehicle travel with respect to fuel cost per mile using the RTEC survey data', Oak Ridge National

Laboratory, Oak Ridge, Tennessee.

Greene, D. L., J. R. Kahn, and R. C. Gibson, (1999b), 'Fuel economy rebound effect for US household

vehicles', Energy Journal, 20(3), 1-31.

Greening, L. A. and D. L. Greene, (1998), 'Energy use, technical efficiency, and the rebound effect:

a review of the literature', Report to the U.S. Department of Energy, Hagler Bailly and Co., Denver.

Greening, L. A., D. L. Greene, and C. Difiglio, (2000), 'Energy efficiency and consumption - the

rebound effect - a survey', Energy Policy, 28(6-7), 389-401.

Grepperud, S. and I. Rasmussen, (2004), 'A general equilibrium assessment of rebound effects',

Energy Economics, 26(2), 261-82.

S4134 12/10/07 13:57 Page 99

Grubb, M., J. Kohler, and D. Anderson, (2002), 'Induced technical change in energy and

environmental modelling: analytic approaches and policy implications', Annual Review of Energy and

the Environment, 27, 271-308.

Grubb, M. J., (1990), 'Energy efficiency and economic fallacies', Energy Policy, 18(8), 783-85.

Grubb, M. J., (1992), 'Reply to Brookes', Energy Policy, 783-85.

Grubb, M. J., (1995), 'Asymmetrical price elasticities of energy demand', in Global warming and

energy demand, T. Barker, P. Ekins and N. Johnstone eds, Routledge, London and New York.

Guertin, C., S. Kumbhakar, and A. Duraiappah, (2003), 'Determining demand for energy services:

investigating income-driven behaviours', International Institute for Sustainable Development.

Gullickson, W. and M. J. Harper, (1987), 'Multi factor productivity in US manufacturing 1949-1983',

Monthly Labour Review, (October 18-28).

Haas, R. and P. Biermayr, (2000), 'The rebound effect for space heating - Empirical evidence from

Austria', Energy Policy, 28(6-7), 403-10.

Haas, R. and L. Schipper, (1998), 'Residential energy demand in OECD-countries and the role of

irreversible efficiency improvements', Energy Economics, 20(4), 421-42.

Hanley, M., J. M. Dargay, and P. B. Goodwin, (2002), 'Review of income and price elasticities in the

demand for road traffic', Final report to the DTLR under Contract number PPAD 9/65/93, ESRC

Transport Studies Unit, University College London, London.

Hanley, N., P. G. McGregor, J. K. Swales, and K. Turner, (2005), 'Do increases in resource

productivity improve environmental quality? Theory and evidence on rebound and backfire effects

from an energy-economy-environment regional computable general equilibrium model of Scotland',

Department of Economics, University of Strathclyde, Strathclyde.

Harrison, D., G. Leonard, B. Reddy, D. Radov, P. Klevnäs, J. Patchett, and P. Reschke, (2005),

'Reviews of studies evaluating the impacts of motor vehicle greenhouse gas emissions regulations

in California', National Economic Research Associates, Boston.

Hartman, R. S., (1988), 'Self-selection bias in the evaluation of voluntary energy conservation

programs', The Review of Economics and Statistics, 70(3), 448-58.

Haughton, J. and S. Sarkar, (1996), 'Gasoline tax as a corrective tax: Estimates for the United

States, 1970-1991', Energy Journal, 17(2), 103-26.

Hausman, J. A., (1979), 'Individual Discount Rates and the Purchase and Utilization of Energy-Using

Durables', Bell Journal of Economics, 10(1), 33-54.

Henly, J., H. Ruderman, and M. D. Levine, (1988), 'Energy savings resulting from the adoption of

more efficient appliances: a follow-up', Energy Journal, 9(2), 163-70.

Herendeen, R., (1978), 'Total energy cost of household consumption in Norway, 1973', Energy, 3(5),

615-30.

Herendeen, R. and J. Tanak, (1976), 'The energy cost of living', Energy, 1(2), 165-78.

Herring, H., (2006), 'Energy efficiency--a critical view', Energy, 31(1), 10-20.

Herring, H. and M. Elliot, (1990), 'Letter to the editor', Energy Policy, 18(8), 786.

Hills, R. L., (1989), Power from Steam, Cambridge University Press, Cambridge.

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

100

S4134 12/10/07 13:57 Page 100

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

101

Hogan, W. W., (1979), 'Capital-energy complementarity in aggregate energy-economic analysis',

Resources and Energy, 2, 201-20.

Hogan, W. W. and D. W. Jorgenson, (1991), 'Productivity trends and the cost of reducing CO2

emissions', The Energy Journal, 12(1), 67-85.

HoL, (2006), 'Second Report: Energy Efficiency, Session 2005-2006', House of Lords Select

Committee on Science and Technology, London.

Hong, N. V., (1983), 'Two measures of aggregate energy production elasticities', The Energy Journal,

4(2), 172-77.

Hong, S. H., T. Oreszeczyn, and I. Ridley, (2006), 'The impact of energy efficient refurbishment on

the space heating fuel consumption in English dwellings', Energy and Buildings, 38(10), 1171-81.

Howarth, R. B., (1997), 'Energy efficiency and economic growth', Contemporary Economic Policy,

15(4), 1.

Hsueh, L.-M. and J. L. Gerner, (1993), 'Effect of Thermal Improvements in Housing on Residential

Energy Demand', Journal of Consumer Affairs, 27(1), 87-105.

IPCC, (2007), 'Climate Change 2007: Mitigation of Climate Change', Working Group III contribution

to the IPCC Fourth Assessment Report, Intergovernmental Panel on Climate Change.

Jalas, M., (2002), 'A time use perspective on the materials intensity of consumption', Ecological

Economics, 41(1), 109-23.

Jevons, W. S., (1865), 'The Coal Question: Can Britain Survive?' in The Coal Question: An Inquiry

Concerning the Progress of the Nation, and the Probable Exhaustion of Our Coal-mines, A.W. Flux

ed, Augustus M. Kelley, New York.

Johansson, O. and L. Schipper, (1997), 'Measuring long-run automobile fuel demand: separate

estimations of vehicle stock, mean fuel intensity, and mean annual driving distance', Journal of

Transport Economics and Policy, 31(3), 277-92.

Jones, C. T., (1993), 'Another look at U.S. passenger vehicle use and the "rebound" effect from

improved fuel efficiency', The Energy Journal, 14(4), 99.

Jones, H. G., (1975), An introduction to modern theories of economic growth, Van Nostrand

Reinhold, London.

Jorgenson, D. and B. Fraumeni, (1981), 'Relative prices and Technical Change', in Modeling and

Measuring Natural Resource Substitution., E.R. Berndt and B.C. Field eds, The MIT Press, Boston.

Jorgenson, D. W., (1984), 'The role of energy in productivity growth.' The Energy Journal, 5(3).

Jorgenson, D. W., (1996), Growth Vol. 2 Energy, the Environment, and Economic Growth, MIT Press,

Boston.

Jorgenson, D. W. and Z. Griliches, (1967), 'The explanation of productivity change', Review of

Economic Studies, 34(3), 249-82.

Junankar, S., O. Lofsnaes, and P. Summerton, (2007), 'MDM-E3: a short technical description',

Working Paper, Cambridge Econometrics.

Kanyama, A. C., R. Engstrom, and R. Kok, (2006), 'Indirect and direct energy requirements of city

households in Sweden: options for reduction, lessons from modelling', Journal of Industrial Ecology,

9(1-2), 221-35.

S4134 12/10/07 13:57 Page 101

Kaufmann, R. K., (1992), 'A biophysical analysis of the energy/real GDP ratio: implications for

substitution and technical change', Ecological Economics, 6(1), 35-56.

Kaufmann, R. K., (1994), 'The relation between marginal product and price in US energy markets :

Implications for climate change policy', Energy Economics, 16(2), 145-58.

Kaufmann, R. K., (2004), 'The mechanisms of autonomous energy efficiency increases: a

cointegration analysis of the US energy/GDP ratio', The Energy Journal, 25(1), 63-86.

Kaufmann, R. K. and I. G. Azary-Lee, (1990), A biophysical analysis of substitution: does

substitution save energy in the US forest products industry?, Ecological economics: its implications

for forest management and research, Proceedings of a Workshop held in St Paul, Minnesota, April

2-6

Kempton, W. and L. Montgomery, (1982), 'Folk quantification of energy', Energy, 7(10), 817-27.

Khazzoom, J. D., (1980), 'Economic implications of mandated efficiency in standards for household

appliances', Energy Journal, 1(4), 21-40.

Klein, Y. L., (1987), 'Residential energy conservation choices of poor households during a period of

rising energy prices', Resources and Energy, 9(4), 363-78.

Klein, Y. L., (1988), 'An econometric model of the joint production and consumption of residential

space heat', Southern Economic Journal, 55(2), 351-59.

Kok, R., R. M. J. Benders, and H. C. Moll, (2006), 'Measuring the environmental load of household

consumption using some methods based on input–output energy analysis: A comparison of methods

and a discussion of results', Energy Policy, 34(17), 2744-61.

Kraft, J. and A. Kraft, (1978), 'On the relationship between energy and GNP', Journal of Energy and

Development, 3(401-3).

Kummel, R., J. Henn, and D. Lindenberger, (2002), 'Capital, labor, energy and creativity: modeling

innovation diffusion', Structural Change and Economic Dynamics, 13(4), 415-33.

Kummel, R., D. Lindenberger, and W. Eichhorn, (2000), 'The productive power of energy and

economic evolution', Indian Journal of Applied Economics, 8, 231-62.

Kummel, R., W. Strassl, A. Gossner, and W. Eichhorn, (1985), 'Technical progress and energy

dependent production functions', Nationalökonomie - Journal of Economics, 3, 285-311.

Laitner, J. A., (2000), 'Energy efficiency: rebounding to a sound analytical perspective', Energy

Policy, 28(6-7), 471-75.

Leach, G., (1975), 'Net energy analysis — is it any use?' Energy Policy, 3(4), 332-44.

Lee, C. C., (2006), 'The causality relationship between energy consumption and GDP in G-11

countries revisited', Energy Policy, 34(9), 1086-93.

Lenzen, M., (1998), 'Energy and greenhouse gas cost of living for Australia during 1993/94', Energy,

23(66), 497-516.

Lenzen, M. and C. Dey, (2000), 'Truncation error in embodied energy analyses of basic iron and steel

products', Energy, 25(6), 577-85.

Lim, H. and C. R. Shumway, (1997), 'Technical change and model specification: US agricultural

production', American Journal of Agricultural Economics, 79, 543-54.

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

102

S4134 12/10/07 13:57 Page 102

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

103

Lipsey, R., G., K. I. Carlaw, and C. T. Bekar, (2005), Economic Transformations: General Purpose

Technologies and Long-term Economic Growth, Oxford University Press, Oxford.

Löschel, A., (2002), 'Technological change in economic models of environmental policy: a survey',

Ecological Economics, 43(2-3), 105-26.

Lovins, A. B., (1998), 'Further comments on Red Herrings', Letter to the New Scientist, No. 2152,

18 September

Lovins, A. B., J. Henly, H. Ruderman, and M. D. Levine, (1988), 'Energy saving resulting from the

adoption of more efficient appliances: another view; a follow-up', The Energy Journal, 9(2), 155.

Lovins, A. B. and L. H. Lovins, (1997), 'Climate: making sense and making money', Rocky Mountain

Institute, Old Snowmass, Colorado.

Meyer, B., (1995), 'Natural and quasi experiments in economics', The Journal of Business and

Economic Statistics, 13(2), 151-60.

Miller, E. M., (1986), 'Cross-sectional and time-series biases in factor demand studies: explaining

energy-capital complementarity', Southern Economic Journal, 52(3), 745-62.

Milne, G. and B. Boardman, (2000), 'Making cold homes warmer: the effect of energy efficiency

improvements in low-income homes', Energy Policy, 28, 411-24.

Mizobuchi, K. I., (2007), 'An empirical evidence of the rebound effect considering capital costs',

Working Paper, Graduate School of Economics, Kobe University, Kobe, Japan.

Nadel, S., (1993), 'The take-back effect - - fact or fiction', U933, ACEEE.

Nesbakken, R., (1999), 'Price sensitivity of residential energy consumption in Norway', Energy

Economics, 21, 493-515.

Norsworthy, J. R., M. J. Harper, and K. Kunz, (1979), 'The slowdown in productivity growth: an

analysis of some contributing factors', Brookings Papers in Economic Activity, 2, 387-41.

Parti, M. and C. Parti, (1980), 'The total and appliance-specific conditional demand for electricity in

the household sector', The Bell Journal of Economics, 11, 309-21.

Patterson, M. G., (1996), 'What is energy efficiency: concepts, indicators and methodological

issues', Energy Policy, 24(5), 377-90.

Poyer, D. A. and M. Williams, (1992), 'Residential energy demand: additional empirical evidence by

minority household type', Energy Economics, 15(2), 93-100.

Puller, S. L. and L. A. Greening, (1999), 'Household adjustment to gasoline price change: an analysis

using 9 years of US survey data', Energy Economics, 21(1), 37-52.

Pye, M. and A. McKane, (1998), Enhancing shareholder value: making a more compelling energy

efficiency case to industry by quantifying non-energy benefits, Proceedings 1999 Summer Study on

Energy Efficiency in Industry, Washington, DC,

Richmond, A. K. and R. K. Kaufmann, (2006a), 'Energy prices and turning points: the relationship

between income and energy use/carbon emissions', The Energy Journal, 27(4), 157-79.

Richmond, A. K. and R. K. Kaufmann, (2006b), 'Is there a turning point in the relationship between

income and energy use and/or carbon emissions?' Ecological Economics, 56(2), 176-89.

Rosen, M. A., (2004), 'Exergy analysis of energy systems', in Encyclopaedia of Energy, C.J.

Cleveland ed, Elsevier Academic Press.

S4134 12/10/07 13:57 Page 103

Rosenberg, N., (1989), Energy Efficient Technologies: Past, Present and Future Perspectives, How

Far Can the World Get on Energy Efficiency Alone?,

Roy, J., (2000), 'The rebound effect: some empirical evidence from India', Energy Policy, 28(6-7),

433-38.

Roy, J., J. Sathaye, A. H. Sanstad, P. Mongia, and K. Schumacher, (1999), 'Productivity trends in

India's energy intensive industries', The Energy Journal, 20(3), 33-61.

Royal Commission on Environmental Pollution, (2007), 'The Urban Environment', London.

Sanders, M. and M. Phillipson, (2006), 'Review of differences between measured and theoretical

energy savings for insulation measures', Centre for Research on Indoor Climate and Health, Glasgow

Caledonian University, Glasgow.

Sanstad, A. H., J. Roy, and S. Sathaye, (2006), 'Estimating energy-augmenting technological change

in developing country industries', Energy Economics, 28, 720-29.

Sartori, I. and A. G. Hestnes, (2007), 'Energy use in the life-cycle of conventional and low-energy

buildings: a review article', Energy and Buildings, 39, 249-57.

Saunders, H. D., (1992), 'The Khazzoom-Brookes postulate and neoclassical growth', The Energy

Journal, 13(4), 131.

Saunders, H. D., (2000a), 'Does predicted rebound depend on distinguishing between energy and

energy services?' Energy Policy, 28(6-7), 497-500.

Saunders, H. D., (2000b), 'A view from the macro side: rebound, backfire, and Khazzoom-Brookes',

Energy Policy, 28(6-7), 439-49.

Saunders, H. D., (2005), 'A calculator for energy consumption changes arising from new

technologies', Topics in Economic Analysis and Policy, 5(1), 1-31.

Saunders, H. D., (2007), 'Fuel conserving (and using) production function', Working Paper, Decision

Processes Incorporated, Danville, CA.

Schipper, L., (2000), 'Editorial', Energy Policy, 28, 6-7.

Schipper, L. and M. Grubb, (2000), 'On the rebound? Feedback between energy intensities and

energy uses in IEA countries', Energy Policy, 28(6-7), 367-88.

Schipper, L., M. Josefina, L. P. Figueroa, and M. Espey, (1993), 'Mind the gap The vicious circle of

measuring automobile fuel use', Energy Policy, 21(12), 1173-90.

Schurr, S., (1982), 'Energy Efficiency and Productive Efficiency: Some Thoughts Based on the

American Experience', Energy Journal, 3(3), 3-14.

Schurr, S. H., (1983), 'Energy abundance and economic progress', Annual Review of Energy,

10(3/4), 111-17.

Schurr, S. H., (1984), 'Energy, economic growth and human welfare', in The Economics of Nuclear

Energy., L. Brookes and H. Motamen eds, Chapman and Hall, London.

Schurr, S. H., (1985), 'Energy conservation and productivity growth : Can we have both?' Energy

Policy, 13(2), 126-32.

Schurr, S. H., B. C. Netschert, V. E. Eliasberg, J. Lerner, and H. H. Landsberg, (1960), 'Energy in the

American Economy', Resources for the Future, John Hopkins University Press, Baltimore.

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

104

S4134 12/10/07 13:57 Page 104

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

105

Schwarz, P. M. and T. N. Taylor, (1995), 'Cold hands, warm hearth? Climate, net takeback, and

household comfort', Energy Journal, 16(1), 41-54.

Small, K. A. and K. Van Dender, (2005), 'A study to evaluate the effect of reduced greenhouse gas

emissions on vehicle miles travelled', Prepared for the State of California Air Resources Board, the

California Environment Protection Agency and the California Energy Commission, Final Report ARB

Contract Number 02-336, Department of Economics, University of California, Irvine.

Söllner, F., (1997), 'A reexamination of the role of thermodynamics for environmental economics',

Ecological Economics, 22(3), 175-201.

Solow, J. L., (1987), 'The capital-energy complementarity debate revisited', The American Economic

Review, 77(4), 605-14.

Sorrell, S., (2007), 'Improving the evidence base for energy policy: The role of systematic reviews',

Energy Policy, 35(3), 1858-71.

Sorrell, S. and J. Dimitropoulos, (2005a), 'An assessment of evidence for a 'rebound effect' from

improvements in energy efficiency: Assessment Protocol', A report for the UK Energy Research

Centre, Sussex Energy Group, SPRU, University of Sussex, Brighton.

Sorrell, S. and J. Dimitropoulos, (2005b), 'An assessment of evidence for rebound effect from

improvements in energy efficiency: Scoping Note', Report for the Technology Policy Assessment

function of the UK Energy Research Centre, SPRU (Science and Technology Policy Research),

University of Sussex, Brighton.

Sorrell, S., E. O. Malley, S. J., and S. Scott, (2004), The Economics of Energy Efficiency: Barriers to

Cost-Effective Investment, Edward Elgar, Cheltenham, UK.

Stern, D. I., (1993), 'Energy and economic growth in the USA : A multivariate approach', Energy

Economics, 15(2), 137-50.

Stern, D. I., (1997), 'Limits to substitution in the irreversibility in production and consumption: a

neoclassical interpretation of ecological economics', Ecological Economics, 21, 197-215.

Stern, D. I., (2000), 'A multivariate cointegration analysis of the role of energy in the US

macroeconomy', Energy Economics, 22, 267-83.

Stern, D. I., (2004a), 'Elasticities of substitution and complementarity', Working Paper in Economics,

No. 0403, Rennsselaer Polytechnic Institute.

Stern, D. I., (2004b), 'The rise and fall of the Environmental Kuznets Curve', World Development,

32(Eight), 1419-39.

Stern, D. I., (2004c), 'The rise and fall of the Environmental Kuznets Curve', World Development,

32(8), 1419-39.

Stern, D. I. and C. J. Cleveland, (2004), 'Energy and economic growth', Rensselaer Working Paper

in Economics, No. 0410, Rensselaer Polytechnic Institute, Troy, NY.

Stern, N., (2007), 'Stern Review: The Economics of Climate Change', HM Treasury, London.

Sue Wing, I. and R. S. Eckaus, (2006), 'The Decline in U.S. Energy Intensity: Its Origins and

Implications for Long-Run CO2 Emission Projections', Working Paper, Kennedy School of

Government, Harvard University, Boston.

Sweeney, J. L., (1984), 'The response of energy demand to higher prices: what have we learned?'

American Economic Review, 74(2), 31-37.

S4134 12/10/07 13:57 Page 105

Taylor, L. D., G. R. Blattenberger, and P. K. Verleger, (1977), 'The residential demand for energy',

EA-235, Electric Power Research Institute, Palo Alto, California.

Temple, J., (2006), 'Aggregate production functions and growth economics', International Review of

Applied Economics, 20(3), 301-17.

Toke, D., (1990), 'Increasing energy supply not inevitable', Energy Policy, 18(7), 671-73.

Toman, M. and B. Jemelkova, (2003), 'Energy and Economic Development: An Assessment of the

State of Knowledge', The Energy Journal, 24(4).

Train, K. E., (1994), 'Estimation of net savings from energy-conservation programs', Energy, 19(4),

423-41.

Vringer, K. and K. Blok, (1995), 'The direct and indirect energy requirements of households in the

Netherlands', Energy Policy, 23(10), 893-910.

Wall, G., (2004), 'Exergy', in Encyclopaedia of Energy: Volume 2, C.J. Cleveland ed, Elsevier

Academic Publishers, Amsterdam, New York.

Webb, M. and D. Pearce, (1975), 'The economics of energy analysis', Energy Policy, 3(4), 318-31.

Welsch, H. and C. Ochsen, (2005), 'The determinants of aggregate energy use in West Germany:

factor substitution, technological change, and trade', Energy Economics, 27, 93-111.

West, S. E., (2004), 'Distributional effects of alternative vehicle pollution control policies', Journal of

Public Economics, 88, 735-57.

Wilson, A. and J. Boehland, (2005), 'Small is Beautiful: U.S. House Size, Resource Use, and the

Environment', Journal of Industrial Ecology7, 9(1-2), 277-28.

Winther, B. N. and A. G. Hestnes, (1999), 'Solar versus Green: the analysis of a Norwegian row

house', Solar Energy, 66(6), 387-93.

Wirl, F., (1997), The economics of conservation programs, Kluwer, Dordrecht.

Worrell, J. A., M. Ruth, H. E. Finman, and J. A. Laitner, (2003), 'Productivity benefits of industrial

energy efficiency measures', Energy, 1081-98.

Xiao, N., J. Zarnikau, and P. Damien, (2007), 'Testing functional forms in energy modeling: An

application of the Bayesian approach to U.S. electricity demand', Energy Economics, 29(2), 158-66.

Yoo, S. H., (2006), 'The causal relationship between electricity consumption and economic growth

in the ASEAN countries', Energy Policy, 34(18), 3573-82.

Zachariadis, T., (2006), 'On the exploration of causal relationships between energy and the

economy', Discussion Paper 2006-05, Department of Economics, University of Cyprus.

Zarnikau, J., (1999), 'Will tomorrow's energy efficiency indices prove useful in economic studies?'

The Energy Journal, 20(3), 139-45.

Zarnikau, J., S. Guermouches, and P. Scmidt, (1996), 'Can different energy resources be added or

compared?' Energy, 21(6), 483-91.

Zein-Elabdin, E. O., (1997), 'Improved stoves in Sub-Saharan Africa: the case of the Sudan', Energy

Economics, 19(4), 465-75.

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

106

S4134 12/10/07 13:57 Page 106

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

107

Annex 1: Project team, expert group and contributors

Project TeamThe assessment was managed by SteveSorrell of the Sussex Energy Group, SPRU,University of Sussex. The contributorswere as follows:

Sussex Energy Group (SEG),University of Sussex

Steve Sorrell,

John Dimitropoulos

Surrey Energy Economics Centre(SEEC), University of Surrey

Lester Hunt

David Broadstock

Fraser of Allander Institute andDepartment Economics, University ofStrathclyde

Grant Allan

Michelle Gilmartin

Peter McGregor

Kim Swales

Karen Turner

ICEPT, Imperial College

Dennis Anderson

Matt Sommerville

Expert GroupThe expert group was chosen for itscombination of expertise on energyefficiency, energy economics and therebound effect. It met twice during thecourse of the project, providing input tothe initial framing of the issues, literaturesearch, synthesis and drafting. Themembers were as follows:

Jim Skea (UKERC);

Horace Herring (Open University);

Hunter Danskin (DEFRA);

Tina Dallman (DEFRA);

Ken Double (Energy Saving Trust);

Lester Hunt (Surrey Energy EconomicsCentre);

Paolo Agnolucci (Policy Studies Institute).

Peer ReviewersManuel Frondel (Zentrum für EuropäischeWirtschaftsforschung, Mannheim)

Karsten Neuhoff (Electricity PolicyResearch Group, Faculty of Economics,University of Cambridge),

Jake Chapman

S4134 12/10/07 13:57 Page 107

The results of the full assessment are contained in five in-depth Technical Reports, asfollows:

Technical Report 1: Evidence from evaluation studies

(Matt Sommerville, Steve Sorrell)

Technical Report 2: Evidence from econometric studies

(Steve Sorrell, John Dimitropoulos)

Technical Report 3: Evidence from elasticity of substitution studies

(David Broadstock, Lester Hunt, Steve Sorrell)

Technical Report 4: Evidence from CGE modeling studies

(Grant Allan, Michelle Gilmartin, Peter McGregor, Kim Swales, Karen Turner)

Technical Report 5: Evidence from energy, productivity and economic growthstudies

(Steve Sorrell, John Dimitropoulos)

In addition, there is a shorter Supplementary Note by Steve Sorrell on the graphicalanalysis of rebound effects. All these reports are available to download from the UKERCwebsite.

Annex 2: Technical and Supplementary Reports

The

Reb

ound

Effe

ct: a

n as

sess

men

t of

the

evi

denc

e fo

r ec

onom

y-w

ide

ener

gy s

avin

gs fr

om im

prov

ed e

nerg

y ef

ficie

ncy

108

S4134 12/10/07 13:57 Page 108

The Rebound Effect:an assessment of the evidence for economy-wide

energy savings from improved energy efficiency

October 2007

Th

e R

eb

ou

nd

Effe

ct:an

asse

ssmen

t of th

e e

vid

en

ce fo

r eco

no

my-w

ide e

nerg

y sa

vin

gs fro

m im

pro

ved

en

erg

y e

fficien

cy

UK

En

erg

y R

ese

arch

Cen

tre


58 Prince’s Gate

Exhibition Road

London SW7 2PG

tel: +44 (0)20 7594 1574

email: [email protected]

www.ukerc.ac.uk


S4134_cover 12/10/07 13:57 Page 1

the rebound effect...the rebound effect: an assessment of the evidence for economy-wide energy...

Documents