Contents lists available at SciVerse ScienceDirect
Energy Economics
j ourna l homepage: www.e lsev ie r .com/ locate /eneco
Saving money vs investing money: Do energy ratings influenceconsumer demand for energy efficient goods?
Luca A. PanzoneSustainable Consumption Institute, and Department of Economics, School of Social Science, University of Manchester (UK), Arthur Lewis building, Oxford Road, Manchester M13 9PL, United Kingdom
Abbreviations: 2SLS, Two-stage least squares; 3SLS, TAlmost Ideal Demand System; BC, Bayonet Cap (lighFluorescent Light bulbs; DECC, Department of Energy anefficiency index; ES, Edison Screw (light bulb fitting); EEnergy Using Product; GLS, General lighting service;Small Edison Screw (light bulb fitting); SUR, Seemingly
Article history:Received 23 August 2012Received in revised form 4 March 2013Accepted 10 March 2013Available online 16 March 2013
JEL classification:C10D12Q41Q48Q55
Keywords:Energy-efficiency gapEnergy-using productsElectricityConsumer behaviourAIDS model
The article analyses economic barriers leading to the energy efficiency gap in the market for energy-usingproducts by observing several million transactions in the UK over two years. The empirical exercise estimatesAIDS models for refrigerators, washing machines, TVs, and light bulbs. Results indicate that market barriersare crucial in the demand for energy efficient options, and consumer response to changes in appliance prices,total expenditures, and energy prices depends on the possibility of behavioural adjustments in consumption.In contrast with the induced innovation hypothesis, current electricity prices can fail to induce innovationbecause of their short-term impact on disposable income, while consumers invest in energy efficiencywhen expecting electricity prices to rise in the future.
Energy conservation originates from the need to preservethe existing stock of non-renewable natural energy resources(e.g. Sutherland, 1996), as well as reducing the economic and politi-cal dependence on these energy goods (Hamilton, 2003; Kilian,2008). Current policies have focused on two main aspects: invest-ment in the production of renewable or less carbon-intensive energysources (Fischer, 2008); and the reduction of energy waste (Linaresand Labandeira, 2010). This second item promotes investments inenergy-efficiency by reducing the amount of energy required toobtain a unit of consumption (e.g. the same amount of light usingless electricity; or travel the same distance with less fuel).
Energy efficiency policy has focused primarily on the supply side,targeting firm investments in efficient technology and imposingproduction standards (e.g. emission standards for cars) (Gillinghamet al., 2009). Efforts to improve household energy consumption
have gained popularity in the last decades, with a focus on environ-mental labelling and more recently on targeted behavioural cam-paigns (e.g. smart meters, carbon calculators) (see e.g. Allcott andMullainathan, 2010). The shift in focus to household consumption isdue to its crucial role in achieving carbon reduction and energyconservation targets in developed economies (Dietz et al., 2009;Vandenbergh and Steinemann, 2007): in the UK, households emitaround 32% of total CO2 (2008 estimates) and consume around 29%of total energy and 38% of electricity (2009 estimates).1 By investing inexisting energy-efficient technology, this sector has the potential forrapid and large reductions in carbon emissions (Dietz et al., 2009),counting on a double incentive to spend: environmental improvements(a social benefit); and monetary savings from reduced waste andcheaper energy consumption (a private benefit) (Gillingham et al.,2009).
Despite the theoretical advantage of purchasing energy efficientappliances, the difference between observed and optimal levels of en-ergy use remains significant (Allcott and Greenstone, 2012; Brennan,2011), a phenomenon known as “energy-efficiency gap” (Gillinghamet al., 2009; Jaffe and Stavins, 1994; Jaffe et al., 2004). According tomodels of rational consumer behaviour, a household is always
1 Eurostat statistics, available at http://epp.eurostat.ec.europa.eu/portal/page/portal/statistics/search_database, section “Environment and energy”.
Fig. 1. Household electricity consumption by area of consumption (2009).Source: DECC (2011).
Light17.15%
Cold16.83%
Wet17.22%
Consumer Electronics
25.11%
Home Computing
7.83%
Cooking15.85%
Fig. 2. Household electricity consumption ('000 tonnes of oil equivalent) by category ofappliance (2010).Source: DECC (2011).
52 L.A. Panzone / Energy Economics 38 (2013) 51–63
expected to invest (i.e. spend) money2 on a new technology whenev-er it provides positive long-term benefits. A violation of this principleis considered a counterintuitive violation of the law of demand: con-sumers who would benefit from investments in energy efficiency do notinvest, causing both a private loss (missed savings, an internality3) and apublic loss (overconsumption of energy, an externality). The energy-efficiency gap could still be a rational behaviour causing no private orpublic loss (Jaffe et al., 2004): costs and benefits of the investmentmight go to different agents (the owner-renter problem, see Davis,2011); and lack of capital might discourage a motivated investor. Thegap could also arise when an agent invests excessively in efficiency, dueto the desire to contribute to a public good even at an economic loss.4
This article advances existing empirical research on the energy ef-ficiency gap by exploring the market for Energy Using Products(EUPs) in the UK. The focus on the discussion is on four goods: lightbulbs, a commodity with limited impact on the budget constraint;refrigerators, a white good with limited substitution in the householdproduction function; TV,5 an appliance used for leisure; and (clothes)washing machines,6 a white good with direct substitutes in the house-hold production function (e.g. laundrettes and laundry shops). Therelevance of these items is depicted in Figs. 1–3: appliances and light-ing account for three quarters of energy consumed within UK house-holds (Fig. 1). Here, consumer electronics, wet appliances, lighting,and cold storage facilities are the four areas with the highest patternsof consumption (Fig. 2), and within each subcategory the articlesfocus on the item consuming the most electricity7 (Fig. 3).
From a behavioural perspective, these four products differ sub-stantially. Refrigerators represent a significant part of the energybill, and are the category with the lowest possibility of behaviouraladjustment because they require continuous electricity consumption.Here, changes in consumer behaviour are limited to substitution to
2 The Merriam-Webster dictionary defines investment as the “Process of exchangingincome for an asset that is expected to produce earnings at a later time” (http://www.merriam-webster.com/dictionary/investment). The present purchase of an energy effi-cient appliance corresponds to the payment of an extra amount of income equal to theprice premium that will produce future savings, hence the use of the term “investment”,as also consistent with the literature (e.g. Jaffe and Stavins, 1994; Sutherland, 1996).The use of this term differs from the usual macroeconomic discussion of investment,which refers to assets that can appreciate in the future and generate benefits upon resale.
3 An internality is defined as a behaviour that imposes extra costs on the agentwho is re-sponsible for it (Herrnstein et al., 1993). In the case of the energy-efficiencygap, the forgoneinvestment imposes extra long-term monetary costs to households because the decision-making process does not fully account for future costs and benefits (Jaffe et al., 2004).
4 I am thankful to an anonymous referee for suggesting this point.5 According to the 2009 Family Spending survey of the ONS, in the UK 97% of house-
holds own a TV.6 According to the 2009 Family Spending survey of the ONS, in the UK 96% of house-
holds own a washing machine.7 Washing machines have been preferred to tumble driers because they are seg-
mented by efficiency class.
other storage type (e.g. canned food); however, because runningcosts are given once the refrigerator is connected, unit costs of storagedecline only with the amount of food stored (i.e. usage is cheapestwhen the refrigerator is full). The purchase of energy efficient lightbulbs is instead partially unrelated to their consumption, becauseitems are often stored for perspective use. Light bulbs usage is alsodiscontinuous and influenced by seasonal factors (e.g. seasonallight). Finally, TVs and washing machines only require discontinuous,albeit periodic, usage and their purchase often accommodates personalneeds (e.g. limited space for washers; or an ornamental TV set). Con-versely, investments in energy efficient EUPs provide long-term reduc-tions in energy consumption with no need for behavioural change.
The provision of energy-efficiency ratings on the labels of EUPs(e.g. class A refrigerator) aims at achieving a change in consumershopping behaviour by making the environmental outcome ofchoices salient and reducing asymmetric information (e.g. Mills andSchleich, 2010). Previous research established that the resistance ofthe gap in the presence of energy ratings can be attributed to marketbarriers, the focus of this article, as well as psychological barriers suchas performance uncertainty and loss aversion (Greene, 2011; Jaffe etal., 2004). Among market barriers, energy and technology pricesplay a central role in household investments in efficient EUPs togeth-er with perceived discount rates, which however are not discussed inthis work.8 While these factors do not constitute market failures, theirrole is mediated by an energy label: the moral duty of environmentalpreservation justifies the investment in an efficient EUP on socialgrounds despite a short-run reduction in disposable income.
The price of EUPs is an important barrier to energy efficiency be-cause it represents the fixed costs of an investment. The purchase ofa new appliance requires a substantial sum of money per se, andenergy-efficient options demand a further price premium (Dale andFujita, 2008; Galarraga et al., 2011a). The same argument applies tolight bulbs, despite representing a relatively small expense (seeSection 3). Moreover, EUPs are characterised by a fast technologicaldevelopment, which induces consumers to expect a fast depreciationof the money they spend (Dubin and McFadden, 1984; Hausman,1979) and imperfections in the functioning of new technologies(Mick and Fournier, 1998). As a result, consumers tend to delay thereplacement of EUPs until unavoidable (Galarraga et al., 2011b;Young, 2008). The overall response to a price change would then de-pend on consumer perception of energy efficiency: efficient EUPswould be expected to be price-sensitive because efficiency is not a ne-cessity feature; however, efficient EUPs tend to have less luxury fea-tures (e.g. refrigerators with no ice dispensers; or washing machineswith no drier), making these products less responsive to price changes.
8 The implications of the absence of this factor on the results are presented at theend of the discussion section.
Fig. 3. Household electricity consumption ('000 tonnes of oil equivalent) by subcategory of appliance (2010).Source: DECC (2011).
53L.A. Panzone / Energy Economics 38 (2013) 51–63
Energy prices are often considered the most important driver ofprivate investments in energy-efficiency, a principle known as the“induced innovation hypothesis” (see Newell et al., 1999). Accordingto this principle, rational agents invest in efficient EUPs when energyis expensive to reduce its running costs (Gillingham et al., 2009;Linares and Labandeira, 2010), a hypothesis supported by evidencefrom the industry sector (Popp, 2002). However, household energyconsumption is inelastic in the short-run (e.g. Reiss and White, 2005)and declines moderately in the presence of price shocks (Reiss andWhite, 2008); moreover, price hikes can cause a short-term reductionon disposable income that discourages investments in energy efficien-cy. Nevertheless, the long-term benefit of investments in efficiencywhen (present and future) energy prices are high represents an incen-tive for consumers to purchase efficient EUPs (Jaffe and Stavins, 1994).Higher sensitivity would then be expected for EUPs requiring continu-ous usage. While a positive link between energy price and ownershipof efficient EUPs appears in stated adoption data (Mills and Schleich,2010), no evidence is available for market data.
Despite its relevance, there is little empirical understanding on theeconomic context where energy-efficiency is traded, particularly onthe demand side. In the absence of estimated demand elasticities forenergy-efficient EUPs from revealed preference data, price interven-tion has been frequently advocated to reduce the energy-efficiencygap in these markets (e.g. Galarraga et al., 2011a; Markandya et al.,2009; Metcalf and Weisbach, 2009; Tietenberg, 2009). However, itis unclear how responsive these markets are to price incentives and
disincentive, and this article addresses this gap of the literature. Pre-vious research addressed the problem estimating equilibrium priceelasticities (e.g. Carman, 1972; Dale and Fujita, 2008), often for awhole category. Works closer to the aims of this article are those ofDale and Fujita (2008), which estimate price elasticities by energy-efficiency class with no structural identification of demand; andGalarraga et al. (2011a), who estimate the implicit demand for energyefficiency in the market for washing machine.
The article estimates demand parameters for different energy classesfrom equilibrium quantity and sales using an Almost Ideal Demand Sys-tem (AIDS). Results indicate that economic barriers are an important lim-itation to investments in energy efficiency for EUPs, and the response toprice and income changes depends on the possibility of behavioural ad-aptation at consumption. For instance, the energy efficiency gap in themarket for TV andwashingmachines is likely caused by consumer indif-ference to the information, and an increase in energy prices can be tack-led by making consumption practices more efficient (e.g. washing atlower temperatures; or avoiding TVs on stand-by). Energy efficiencymatters for refrigerators and lighting, where consumers find more diffi-cult to switch to alternative storage or lighting technologies, makingprices themost stringent barrier to investments in thesemarkets. Finally,future electricity prices stimulate the purchase of efficient appliances,while current electricity prices can fail to influence the demand forenergy-efficient EUPs. These findings suggest that while eco-innovationcan drive behavioural change (see Dietz et al., 2009; Ekins, 2010), eco-nomic barriers vary depending on how consumers use EUPs.
Table 1Formula for elasticities and their standard errors in an AIDS model.
9 The test statistic for γ (the same applies for β) is {(γuncorr − γcorr) ⋅ [var(γuncorr) −var(γcorr)]−1 ⋅ (γuncorr − γcorr)}, where var(.) is the variance. The resulting coefficient isχ2-distributed with degrees of freedom equal to the number of endogenous variables.Under the null hypothesis, both estimators are consistent (no endogeneity) (Dhar et al.,2003).
54 L.A. Panzone / Energy Economics 38 (2013) 51–63
The next section outlines the econometric model used in the em-pirical analysis. Demand parameters have been estimated observingmillions of transactions (revealed preferences) in the UK. Such arich dataset, which is described in Section 3, appears to be uncommonin the energy-efficiency literature. Compared to previous research,the article reaches a good compromise between the depth andscope of the analysis: it focuses on four product categories withlarge policy implications in terms of household energy consumption;but in doing so it simplifies the estimation procedure by limiting thedimension of product differentiation to the energy efficiency classonly. The estimation procedure moderates the absence of productheterogeneity by correcting for endogeneity of price and expendi-tures. Section 4 describes the results, whose implications are debatedin Section 5. Section 6 concludes.
2. The econometric model
Demand parameters for appliance by energy-efficiency class, havebeen estimated using an AIDS model (Deaton and Muellbauer, 1980).In a market with N different energy efficiency classes, the expenditureshare function for products in class i at time t can be written in the form
wit ¼ αit þ∑jγij log pjt
� �þ βi⋅log Xt=Ptð Þ þ eit: ð1Þ
In Eq. (1), w indicates expenditure share, p refers to the averageprice paid for every energy class, X is total expenditures (and X/P isreal expenditures), while e is the statistical error. The price index Ptin Eq. (1) corresponds to
log Ptð Þ ¼ α0 þ∑iαi log pitð Þ þ 1
2∑j∑iγij log pitð Þ log pjt
� �ð2Þ
which in the empirical literature (e.g. Blanciforti and Green, 1983;Chalfant, 1987; Green and Alston, 1990) is replaced by the simplerStone's index
Pt� ¼ ∑
iwit log pitð Þ: ð3Þ
The approximation in Eq. (3), adopted in this empirical analysis,gives this model the name of Linear Approximated AIDS (LA/AIDS)(Blanciforti and Green, 1983; Green and Alston, 1990).
The intercept of Eq. (1) can be redefined as αit ¼ αi þ δi⋅ct þϕi⋅ℓtþk to allow for the influence of time-varying parameters ct andfuture electricity prices ℓtþk acting as intercept shifters (Blancifortiand Green, 1983; Dhar et al., 2003; Verbeke and Ward, 2001). Specif-ically,ℓtþk refers to expected UK electricity prices k time periods afterthe purchase (see Section 4 and Appendix 1 for details), while ctincludes a time trend and category-specific time-varying covariatesdescribed in Section 4. The final estimated expenditure share functioncorresponds to
wit ¼ αi þ δi⋅ct þ ϕi⋅ℓtþk þ∑jγij log pjt
� �þ βi⋅ log Xt=P
�t
� �þ eit ð4Þ
where eit is assumed to be normally distributed and affected by serialautocorrelation in the form
eit ¼ ρ·ei;t−1 þ υit : ð5Þ
Eq. (4) corresponds to a systemofN simultaneous demand equations,
whereXN
i¼1
wit ¼ 1. Identification of demand parameters is achieved by
imposing: the adding-up condition,XN
i¼1
αi ¼ 1,XN
i¼1
γij ¼ 0,XN
i¼1
βi ¼ 0,
XN
i¼1
δi ¼ 0 andXN
i¼1
ϕi ¼ 0; the symmetry condition γij = γji; and the
homogeneity condition ∑jγij ¼ 0. Demand elasticities for each category
i are derived from the estimated coefficients using the formulas reportedin Table 1 (see Chalfant, 1987; Filippini, 1995; Green and Alston, 1990;Larivière et al., 2000).
A known limitation of the AIDS model is the inability to deal withproduct heterogeneity (Dhar et al., 2003; Nevo, 2000), which insteadcharacterises the four markets in analysis. As a consequence, theresiduals of Eq. (4) may contain unobservable product characteristicscorrelated with prices pj. Similarly, expenditures X could be correlat-ed with unobservable income shocks in the residuals. To test and cor-rect for the possible endogeneity of pj and X, the article implementsan instrumental variable approach consisting of two stages. The firststage estimates a set of reduced-form instrumental equations (seeDhar et al., 2003), one for each price variable as
ln pjt� �
¼ φ0 þ φ1⋅ht þ ujt ð6Þ
and one for expenditures
ln Xtð Þ ¼ λ0 þ λ1⋅zt þ εt ð7Þ
where u and ε are autocorrelated errors. The variable h indicates ex-ogenous price shifters, while z refers to exogenous expenditureshifters, both expected to have no direct influence on w.
The second stage replaces the observed values of ln(pjt) and Xt inEq. (4) with the fitted values from Eqs. (6) and (7). Rank and orderconditions for identification require at least one exogenous instru-ment for each endogenous variable to be significantly different fromzero in all reduced form equations. Once instrumental equations areestimated, endogeneity can be checked using a Hausman–Wu test(Dhar et al., 2003), which compares endogeneity-corrected coeffi-cients, γcorr and βcorr, with the same uncorrected coefficients, γuncorr
and βuncorr.9 In models with more instruments than endogenous vari-ables, the influence of h and z on market share can be observed test-ing for overidentifying restrictions (Appendix 2). Because this articleuses different datasets, instruments differ across data source and aredescribed in Section 4.
3. Data and descriptive statistics
Revealed preference data consists of total weekly sales and expen-ditures by energy efficiency class for a two-year period for four EUPs:light bulbs, televisions, washing machines, refrigerators. Data onwashing machines and refrigerators were supplied by the market re-search company GfK, and represent live sales at a large sample of UKretailers (among others, Comet, Argos, John Lewis, B&Q, and IKEA)over 100 weeks. The sample only includes the top 50 products soldin each week nationwide. Data on TV and light bulbs was obtained
Representativeness around 23% around 37% around 10% around 7%
a This value represents 99.69% of total units sold available once excluding the cate-gories with gaps.
b This value represents 98.91% of total units sold available once excluding the cate-gories with gaps.
55L.A. Panzone / Energy Economics 38 (2013) 51–63
from Dunnhumby Ltd., and contains purchases recorded in all UKTesco supermarkets over 104 weeks. The sample, summarised inTable 2, accounts for more than a million transactions for each EUPfrom a heterogeneous sample of consumers, and covers from 7% to37% of the whole UK market.
Both data sources have a potential selection problem. GfK dataonly includes consumers who shop in stores supplying informationto the market research company. The sample also excludes productswith low market share. Similarly, Dunnhumby data only accountsfor consumers shopping at Tesco, which only caters for a portion ofUK consumers who might not always use their loyalty card. Neverthe-less, the profile of GfK customers is considered representative of themajority of consumers in mainland Britain.10 Likewise, Dunnhumbycollects information on approximately 17 million active shoppers ayear in a retailer (Tesco)with amarket share of around 30.7% and locat-ed in every UK postal area.11 Due to the strategic importance of theirsample, the owners of the data (the retailers) tend to be parsimoniousin the information they provide to third parties on their customerbase, making a clear comparison of the samples with the UK populationnot possible.
All products in these datasets have been aggregated according tothe energy efficiency rating appearing on their label, which dependson a EUP-specific Energy Efficiency Index (EEI). For refrigerators,the EEI is calculated as annual power consumption relative to a refer-ence point, given its storage volume (EU, 2010a). The EEI for washingmachines, originally conceived as the energy needed for a standardwash, currently measures annual electricity consumption given loadsize (EU, 2010b). The EEI of televisions measures electricity consump-tion relative to a reference given the area of its screen (EU, 2010c);however, TVs were not labelled in the period analysed, despite theEEI being available (see Section 4.3). Importantly, for refrigerators,washing machines and TV the rating is assigned within size catego-ries: two appliances with significantly different size (e.g. a large anda small washing machine) can receive the same rating (e.g. A+)while using significantly different amounts of electricity.12 Finally, theenergy class of light bulbs depends on its luminous flux (in lumens)and its power consumption (in watts) (EU, 1998).
Before proceeding, it is worth exploring the extent of the energy-efficiency gap in the four markets considered. While an accurate esti-mate of the gap is not straightforward (see Allcott and Greenstone,2012), a simple observation of market shares of different energy clas-ses (Figs. 4–5) indicates that three of the four markets present a sub-stantial dominance of relatively inefficient (yet relatively efficient)
10 I am thankful to Anthony Williams (GfK) for this information.11 I am thankful to Chris Gartside (Dunnhumby) for this information.12 See http://www.energysavingtrust.org.uk/Electricals/Products-and-appliances#size.
EUPs. For instance, adoption of the most efficient refrigerator class(A++) was in the range 0–1.17% across the two-year period, whilethe second most efficient alternative (A+) reached no more than15.58% (Fig. 4a). Similarly, most efficient washing machines (A++)did not pass a mere 1.62%, while sales of A+ goods fit the interval1.29%–15.74% (Fig. 4b). High-efficiency televisions reached a maxi-mum market share of 19.35% (Fig. 4c). Finally, efficient light bulbsreached large market shares for all fittings (75%–90%, Fig. 5).
In terms of prices (Tables 3a and 3b), the cost of refrigerators in-creases with energy efficiency, and a similar pattern emerges forA-rated washing machines. B-rated appliances appear to be more ex-pensive than those having A or A+ standards, likely because of a largersize or higher supply costs. A-rated products are always the cheapest,accounting for their large market share. Energy-efficient TVs costmore than twice their non-efficient counterparts, while an environ-mentally friendly light bulb costs around £0.20–£0.25 more than anunfriendly one. While a market premium for energy-efficient productsis consistent with expectations (see e.g. Galarraga et al., 2011a), this isnot a ceteris paribus analysis, and the observed premium could be bun-dling energy efficiencywith other luxury characteristics. Finally, despiteexpectations of a decline in equilibrium prices over time (Dale et al.,2009) prices remained fairly stable in the short period covered for allproducts except light bulbs. The next subsections will estimate anddiscuss demand parameters for the four markets in considerationseparately.
4. Results
To estimate the parameters of AIDS model, total expenditure andtotal quantity sold in a week for each energy-efficiency class havebeen determined as the sum of all products within the class. Averageprices were calculated dividing total expenditure by total quantity.The analysis focuses only on efficiency classes that have sold at leastone unit in each week of the two-year period (98–100% of sales,Table 2). Categories with N > 2 (washingmachines, light bulbs) are es-timated using 3SLS regressions, while those with only two segments(TV, refrigerators) use a 2SLS approach estimated for both classes.Because the presence of identifying restrictions makes the estimationof a system of N simultaneous equations unfeasible (Filippini, 1995;Poi, 2002), the least efficient option is excluded from the SUR stageand all missing coefficients are derived by bootstrapping the adding-up and symmetry conditions. All models correct for serial autocorrela-tion following Chalfant (1987) and Blanciforti and Green (1983).Elasticities follow Table 1, and their standard errors are the observedstandard deviations of 1000 bootstrap replications of the mean acrosstime periods. Finally, Eqs. (6) and (7) have been estimated usingPrais–Winsten regressions to remove serial autocorrelation.
To test for the induced innovation hypothesis, all equations in-clude the monthly residential electricity Retail Price Index 12 monthsafter the observation (January 2005 = 100) from the UK Departmentof Energy and Climate Change.13 A one year lead prices was deter-mined through a k-order autoregressive model (Appendix 1). Toadd robustness to the results of the induced innovation hypothesis,all regressions include a logarithmic time trend that captures the pos-sible changes in consumer tastes toward energy efficient productsunrelated to energy prices that could have been caused by the in-crease in electricity prices after April 2011 (Fig. A.1). All equationsalso contain a set of monthly dummies (baseline: January) to adjustfor seasonality (Blundell and Robin, 2000). Demand models for appli-ances also correct for seasonally adjusted monthly unsecured grosslending to consumers (million GBP, from the Bank of England14) in
13 Data is available at http://www.decc.gov.uk/en/content/cms/statistics/energy_stats/prices/prices.aspx.14 Data, along with a description, can be found at http://www.bankofengland.co.uk/mfsd/iadb/notesiadb/ltoi.htm.
Fig. 4. Market share (quantity) of EUPs by energy efficiency rating.
56 L.A. Panzone / Energy Economics 38 (2013) 51–63
the month of the observation to account for incentives reducing thefixed costs of technology investments. Finally, the model for lightbulbs adjusts for the number of hours of sunshine in the correspond-ing month, which influences the replacement rate (Richardson et al.,2009; Stokes et al., 2004) (MetOffice data15).
The treatment of endogeneity requires the definition of a set of in-strumental variables. The vector h includes price shifters not directlyrelated to market shares (Dhar et al., 2003). In the period in analysisVAT changed twice, moving from 15% to 17.5% on 01/01/2010, and to20% on 04/01/2011,16 and these changes have been coded as a set ofdummies for all appliances (they showed no impact on light bulbprices). Clubcard data also include the percentage of sales (units)sold at a discount in each week for each category. The vector z con-tains variables that increase total expenditures with no effect on mar-ket share (Dhar et al., 2003). Clubcard data use the total number of UKcustomers in a given week; GfK data on refrigerators adjust for totalexpenditures on washing machines (GBP), while washing machinesadjust for total expenditures on refrigerators (GBP). Expendituremodels also adjust for a logarithmic time trend and prices (Blundelland Robin, 2000). Tables 4a, 4b, and 4c present the results ofbootstrapped Hausman endogeneity tests. Results from all instru-mental equations are reported in Appendix 2 together the corre-sponding J-test for overidentifying restriction when applicable (seeAppendix 2).
4.1. Refrigerators
Table 5 presents the results and the estimated elasticities of theAIDS for refrigerators. The Hausman test in Table 4a provides no
15 Data can be found at http://www.metoffice.gov.uk/weather/uk/climate.html.16 See http://www.hmrc.gov.uk/vat/forms-rates/rates/rate-changes.htm.
evidence of endogeneity of both prices and expenditures, whichhave not been instrumented. Results indicate that the price of goodsis crucial in the choice of energy efficient refrigerators, and consumersview the demand of different energy classes as (weak) complementsrather than substitutes. Inefficient (class A) goods are more price sensi-tive than efficient (A+) goods, with own-price elasticities εA = −0.01and εA+ = −0.95. Despite the better environmental performance, A+refrigerators are also relative necessities (ηA+ b 1). Real expendituresdiscourage the purchase of A+ refrigerators, while the availability ofconsumer credit has a positive influence on the demand for efficient op-tions. Finally, future energy prices encourage the purchase of the mostefficient class to the detriment of inefficient options.
These results suggest thatfixed costs of efficient technology adoptionare a relevant barrier in the purchase of efficient refrigerators, but this isnot a problem of disposable income. This result is consistent with thehigh discount rate observed in this market (Meier and Whittier,1983): consumers might see price as a limitation because of the fastdepreciation of new technologies. At the same time, consumers seemto recognise and value the presence of environmental information(Anderson and Claxton, 1982; Mills and Schleich, 2010; Revelt andTrain, 1998), recognising the segmentation by energy-efficiency class.Interestingly, consumers see efficient goods as necessities, and respondlittle when their price changes, but switch to inefficient options whenmore income is available. The missing availability of product attributesprevents the possibility of an accurate explanation of this result, butthe orthogonality between energy efficiency and volume17 and theabsence of price endogeneity suggest that this is unlikely to be a majorlimitation. Instead, these results suggest that consumers consider energyefficiency as a necessity when choosing refrigerators.
17 See http://www.energysavingtrust.org.uk/Electricals/Products-and-appliances#kitchen.
Fig. 5. Market share (quantity) of light bulbs by energy classes and fitting.
57L.A. Panzone / Energy Economics 38 (2013) 51–63
4.2. Washing machines
Table 6 presents results and elasticities from the AIDS model forwashing machines. Total expenditures are endogenous (Table 4a),and have been instrumented using Eq. (7). Results indicate thatdemand for all energy-efficiency classes responds little (p > 0.10) toprices, and own-price elasticities increase (in absolute value) with theenergy rating from εB = −0.71 to εA + = −1.67 (as in Galarraga etal., 2011a). A+ and B options are complements, while all other pairssubstitute each other. A+-rated and B-rated goods are perceived asluxury, whereas A-rated machines are relative necessities (η b 1). Asbefore, the absence of price endogeneity and orthogonality between en-ergy class and load size suggests that unobservable product characteris-tics do not fully explain these results, and energy efficiency is very likelya luxury feature for a washing machine. Moreover, consumer creditfavours the market share of A-rated machines, while real expendituressupport A+ options. A significant time trend suggests that learningfavours A+ washers to the detriment of B goods, while bearing no im-pact on A-rated options. Finally, future electricity prices stimulate theshare of A+ washers while reducing that of A-rated goods, and withno influence on B machines.
Table 3aAverage appliance price by energy class — refrigerators and washing machines.
Note: Averages differ in the number of time periods considered in each year. C-ratedwashing machines appear only in one time period, and their value is not reported.
Overall, the market for washing machines appears to be weaklysegmented by energy class (see also Sammer and Wüstenhagen,2006), and preferences for efficient options depend on income (butnot credit) availability and the prospective price of energy. Notethat washing machines are generally replaced sooner than otherEUPs (Young, 2008), and this process might facilitate learning. Thehigh own-price elasticity observed in efficient washers suggests thatconsumers might perceive the environmental performance of thesegoods as a rather unessential feature, and small price increases leadto substantial drops in energy efficiency. While this price responsemight be caused by a motivational crowding-out (e.g. Perino et al.,forthcoming), where consumers refuse to pay for their pro-socialbehaviour, the high expenditure elasticity provides supporting evi-dence that A+ level of efficiency is a luxury.
4.3. Televisions
Table 7 presents results and elasticities of the AIDS model for TVs.Table 4b indicates that prices and expenditures are endogenous andthey have been instrumented. The market for TVs differs from theothers studied in this article because information on the energy per-formance of goods was not available to consumers in the period
Table 3bAverage appliance price by energy class — TV and light bulbs.
Significance is as follows: * = 0.1; ** = 0.05; *** = 0.01. Based on 1000 bootstrapreplications.
Table 4bBootstrapped Hausman test of endogeneity — Televisions.
Equation No-EST EST
d.f. Chi2 Chi2
All 2 25.05*** 25.10***Prices only 1 9.58*** 13.37***ln(Pno-EST) 1 9.58*** 13.37***ln(PEST) 1 9.58*** 13.37***Real Expenditures 1 13.90*** 10.75***
Significance is as follows: * = 0.10; ** = 0.05; *** = 0.01. Based on 1000 bootstrapreplications.
58 L.A. Panzone / Energy Economics 38 (2013) 51–63
analysed. However, the EEI varied across products and segmentationfollowed the recommendation of the Energy Saving Trust18 (EST):televisions recommended by the EST were classified as “EST”, whileall others were classified as “no-EST”. Results indicate that the costof energy efficiency is the most important economic barrier to con-sumers, both in terms of appliance price and income availability.Own-price elasticities are large (>1 in absolute value) and efficientoptions are substantially more responsive to price (εEST = −9.12;εno-EST = −1.65). Inefficient no-EST televisions are luxury goods,while EST options are inferior goods with negative expenditure elas-ticity. Products in different energy classes strongly substitute eachother. Finally, time stimulates consumption of efficient over ineffi-cient options.
Results suggest that EST televisions are viewed as less attractiveoptions compared to no-EST ones, leading to a possible conflict be-tween public (environmental performance) and private benefitsfrom consumption (e.g. large screen size). The absence of an energylabel limits the ability of consumers to consider and value the envi-ronmental attribute in TVs. The problem is exacerbated by the highaverage market price for EST TVs, which give consumers no incentiveto switch despite the relatively high price elasticity of no-EST options.This conflict could originate from technical limitations preventingproducers from bundling energy efficiency with other attributes at-tractive to consumers. As before, the analysis retains limitations of anon-ceteris paribus analysis of the attributes: some TVs possessmany features that could further explain consumer behaviour, butare not incorporated in this analysis.
4.4. Light bulbs
Table 8 presents results and elasticities of the AIDS model for lightbulbs. In the demand system, products have been classified in six cate-gories identifying both energy-efficiency class and fitting. In particular,light bulbs are either relatively efficient Compact Fluorescent Bulbs(CFL), or relatively inefficient General Lighting Service (GLS).19 Byfitting, light bulbs can be classified as Bayonet Cap (BC), Edison Screw(ES), or Small Edison Screw (SES). This double layer of characteristicsaccounts for structural barriers to consumption, where choices areconstrained by the lamp holding the bulb. Endogeneity affects total ex-penditures in three equations and several prices (Table 4c), and allequations have been instrumented. Because the availability of GLS op-tions has been targeted by EU policies, regressions include twodummies capturing EU regulation: the first adjusts for the ban on100 WGLS (01/09/2009–31/08/2010); the second corrects for the addi-tional ban on 75 W GLS (from 01/09/2010).
18 These can be found at http://www.energysavingtrust.org.uk/Consultancy-and-certification/Energy-Saving-Trust-Recommended/Product-certification.19 See http://www.energysavingtrust.org.uk/In-your-home/Lighting/Lighting-productsfor the environmental performance of different products.
Results indicate that prices influence demand primarily whenproducts are within the same energy class or fitting, although ESGLS and BC CFL appear unresponsive to prices. In general, energy in-efficient options appear more responsive to own-price changes thanefficient alternatives, while showing the same expenditure elasticity(just below unity for CFLs; just above unity for GLSs). It is worth not-ing that the response to prices in the period considered might besomehow distorted: in 2009 energy suppliers distributed free CFLlight bulbs to consumers,20 and subsidised CFL retail prices until the31st of March 2011, causing consumers to respond less to pricechanges in the CFL market. Products substitute each other, with theexception of SES options (generally used for desk lamps) that com-plement ES options. Surprisingly, high future electricity prices andthe time trend favour some energy intensive GLS options. Finally,the phase-out of energy-intensive light bulbs led to a significant declineof inefficient BC and GLS options, affecting negatively also the share ofES CFL; the first phase-out stimulate consumption of efficient SES.
4.5. Electricity prices and the induced innovation hypothesis
This section explores specifically on the role of electricity prices onthe demand for EUPs, which has been commented swiftly inSections 4.1–4.4. Tables 5–8 used future prices to retain consistencywith the induced innovation hypothesis. The previous set of resultsis now extended by re-estimating the same models using electricityprice in the time window t − 12 to t + 12,21 including prices oneat a time to avoid collinearity. Table 9 reports the elasticity of currentand future electricity prices. High current electricity prices favour theconsumption of efficient refrigerators and televisions, as well as inef-ficient washing machines and light bulbs (excluding SES fitting). Con-versely, high future prices increase the demand for all efficientoptions, with the exception of ES and SES light bulbs. Fig. 6 showsall estimated elasticities in the 2-year interval. Future and past elec-tricity prices are very important for the purchase of efficient whitegoods, particularly in the interval of ±6 months. When purchasing ESTtelevisions consumers pay attention mainly to very recent price hikes,responding positively also to prices 6 months or more ahead. High en-ergy prices instead favour inefficient (and cheaper) GLS light bulbs.
These results indicate that current electricity prices can be ineffec-tive in driving the demand for energy-efficient EUPs, while consumersprimarily focus on future prices. The overall picture suggested bythese findings is that in some cases in the short-run consumers mightfocus more on the impact of energy prices on disposable income,adopting a myopic behaviour that leads them to prefer cheaper andmore inefficient energy classes. This point is particularly evident ob-serving the adverse impact of electricity prices on efficient light bulbs.Consequently, while the induced innovation hypothesis suggests that
20 In 2009, utility companies distributed an estimated 200 million light bulbs free toconsumers. See e.g. http://www.telegraph.co.uk/finance/personalfinance/consumertips/household-bills/7511375/Free-energy-saving-bulbs-cost-45.html and http://www.which.co.uk/news/2010/03/shadow-cast-on-free-energy-saving-light-bulbs-207174/.21 I am grateful to an anonymous referee for suggesting the possibility of exploringpast as well as future energy prices.
Significance is as follows: * = 0.10; ** = 0.05; *** = 0.01. Based on 1000 bootstrapreplications.
59L.A. Panzone / Energy Economics 38 (2013) 51–63
energy prices are an important determinant of technological change(Newell et al., 1999; Popp, 2002), consumers use them in the choiceof efficient appliances particularly when expected to increase. This hy-pothesis may not hold when current prices are high: investments inenergy-efficient appliances require fixed costs that decrease disposableincome in the short-run, and grant benefits from lower running costs ofenergy only in the long-run.
5. Discussion
Despite the important environmental andwelfare contributions thatcould come from the change of the current appliance stock to more ef-ficient technology, consumers do not seem to invest sufficiently in ener-gy efficiency, a phenomenon known as the energy-efficiency gap. Thisarticle presents demand estimates for energy-efficient EUPs, reportingdemand parameters for refrigerators, washing machines, TVs, andlight bulbs obtained from four AIDS models. Using revealed preferencesdata from severalmillion transactions in theUK, the article observed theimpact of market barriers such as prices, real expenditures, consumercredit, and electricity prices on the demand for energy-efficient appli-ances. The model tests for the endogeneity of prices and expenditures,correcting the problem when detected.
Results indicate that the way consumers use EUPs in their life de-termines the way consumer respond to market incentives (see alsoWarde, 2005; Shen and Saijo, 2009). For instance, refrigerators
Table 5Parameter estimation of the AIDS for refrigerators.
N = 100. Estimation: constrained linear regressions. Significance is as follows: * = 0.10;** = 0.05; *** = 0.01.
a Bootstrapped standard errors (1000 replications).
require almost continuous use during their lifetime, and consumersconsider paying more to gain long-term economic benefits, makingenergy efficiency a necessity. Increasing energy prices motivate in-vestments in efficient refrigerators. When using washing machinesor televisions, consumers can instead reduce their energy bills bydecreasing the energy consumed per unit of output through be-havioural adjustment. For instance, energy efficiency in washingcan be improved by choosing shorter programmes, lowering watertemperatures, increasing load size, or increasing consumption ofwashing-related services (e.g. laundry-shops) (see e.g. Davis, 2008),making efficiency a luxury feature. For TVs, efficiency of consumptioncan be improved by reducing usage or energy consumed during rest(e.g. removing the standby option; e.g. Duarte and Schaeffer, 2010;Brenčič1 and Young, 2009). Inefficient appliances can then be usedfor conspicuous consumption by signalling high disposable incomethrough energy waste. Consequently, economic barriers to energy ef-ficiency seem to be important concerns for appliances with limitedpossibility for behavioural adaptation.
The case of light bulbs requires separate considerations. Heremarketincentives occasionallywork in thewrong direction, and high electricityprices increase demand for inefficient SES and ES light bulbs to the det-riment of their efficient counterparts. The counterintuitive impact ofelectricity prices suggests that disposable income is an important barri-er to investments in energy-efficiency, and an increase in energy pricesstimulates savings. This result is consistent with behavioural adjust-ments in electricity consumption: in the immediacy of a price hike, con-sumers are more likely to react by reducing electricity consumptionthan invest in energy efficiency (Reiss and White, 2008). Public policyphasing out inefficient options has successfully reduced consumptionof GLS options (the final phase-out occurred in September 2012), andthe high elasticity of substitution between GLS and CFL suggests priceintervention could have also been effective. Nevertheless, the purchaseof inefficient light bulbs delays a decline in energy consumption be-cause an increase in the stock of inefficient goods postpones improve-ments in energy intensity.
In detail, the article provides four main findings contributing toexisting economic research on energy-efficiency. The first finding isthat technology costs (own-prices) influence the decision on the en-ergy class of refrigerators, light bulbs and TV, while economic wealth(real expenditures) directs the choice of washing machines, and(negatively) TV and refrigerators. Hence, consumers view efficient re-frigerators and light bulbs as relatively price-unresponsive necessi-ties. Energy efficiency seems to matter less for washing machines,where consumer view efficient options as luxury and respond consid-erably to own-price changes; and for televisions, where consumersswitch significantly away from efficient options when price increasesdespite being necessities. As a consequence of this result, price-basedpolicy intervention is going to be more effective in markets whereprice stimulates efficiency investments: it should target price-elasticoptions in each category, subsidising elastic efficient EUPs (washingmachines and TV) or taxing elastic inefficient EUPs (light bulbs andrefrigerators). Needless to say, goods in these markets are highly dif-ferentiated and this finding is limited by the absence of a ceterisparibus analysis.
The low interest shown to the energy efficiency label in somemar-kets suggests the presence of important non-market barriers beyondthose commonly studied. For instance, consumers might selectivelyscreen out the information on efficiency in some markets to limitproblems of bounded rationality (e.g. Kahneman, 2003) and informa-tion overload (Fasolo et al., 2007). This idea is supported in a marketwith positive elasticity of substitution (TVs, washing machines andlight bulbs): individuals could be shopping for a EUP consideringefficient and inefficient options separately. Similarly, social and be-havioural barriers could limit the adoption of energy efficient tech-nology by households (see also Allcott, 2011). For instance, theseparation between costs and benefits of investments in energy
Table 6Parameter estimation of the AIDS for washing machines.
A 1.8100*** 0.0369B −0.7956*** 0.0580 0.7633*** 0.0082
Correlation of residuals A+ 1.0000A −0.8548 1.0000
RMSE 0.0171 0.0173Chi2 92.47*** 58.72***
N = 100. Estimation: Seemingly unrelated regressions. Breusch–Pagan test: Chi2(10) = 73.068***. Significance is as follows: * = 0.10; ** = 0.05; *** = 0.01.a Bootstrapped standard errors (1000 replications).
60 L.A. Panzone / Energy Economics 38 (2013) 51–63
efficiency, i.e. the “renter–owner problem” (Davis, 2011) could explainthe low sensitivity to changes in the price of efficient refrigerator and in-efficient washing machine. Behavioural economics could shed somelight on non-monetary barriers to adoption of energy efficient technol-ogy (Allcott and Greenstone, 2012; Allcott and Mullainathan, 2010).
A second finding is that current energy prices do not always driveinnovation, at least not as much as generally conjectured from the in-duced innovation hypothesis. Results indicate that high electricityprices in the future favour the demand for efficient EUPs, while cur-rent electricity prices can discourage investments in energy efficien-cy. High prices always deter consumption of efficient light bulbs. Acomparison of the energy price elasticity before and after the time
Table 7Parameter estimation of the AIDS for televisions.
N = 104. Estimation: constrained linear regressions. Significance is as follows: * =0.10; ** = 0.05; *** = 0.01.
a Bootstrapped standard errors (1000 replications).b Estimates based on periods where wEST > 0(N = 98).
of the observation suggests that demand for efficient white goods issensitive to past and future prices, efficient TV are influenced by pres-ent energy prices, while high energy prices stimulate consumption ofcheap and inefficient GLS light bulbs. The importance of these resultsstems from the realisation that the literature has mainly focusedon the relation between energy prices and patents in efficient tech-nology, hence on the supply side of innovation (e.g. Newell et al.,1999; Popp, 2002). This article focuses on the demand side and ob-serves similar patterns for household behaviour.
This finding is likely related to the way information on energyprice is conveyed to consumers. In fact, energy costs and applianceconsumption is disconnected: most consumers receive delayed feed-back on consumption through a quarterly electricity bill. The ob-served peaks and trends for all EUPs seem to support this idea:consumers show high or increasing sensitivity to the expected energyprice in the following bill, around 4–6 months after purchase. The lowrelevance of energy price on the demand of inefficient washing ma-chine and televisions could instead be caused by the tendency of con-sumers to underestimate energy consumption of infrequently usedappliances. Finally, the negative impact of energy prices on lightbulbs might indicate that consumers do not view efficient lightbulbs as key to reduce energy consumption, possibly expectingmore substantial results from investments on appliances.
Another important consequence of the relation between electrici-ty prices and energy efficiency is that, contrary to findings on the sup-ply side, energy prices can have a substantial negative short-runimpact on the energy efficiency gap. In fact, energy demand is inelas-tic and an increase in current energy prices reduces purchasingpower. Rather than invest, consumers then adjust by reducing con-sumption of both electricity (Reiss and White, 2008) and its comple-ments, such as appliances (Edelstein and Kilian, 2009). Results in thisarticle suggest that consumers also respond by delaying investmentin energy efficient EUPs, unless prices are expected to rise. At thesame time, consumers are likely to adjust their behaviour, changingthe way they use appliances. This short-run response could be causedby a myopic behaviour, whereby consumers show concern on theshort-term loss caused by energy prices rather than the long-term
22 Despite being unable to provide exact figures due to a confidentiality agreement,the data used here cost around 30–40% of equivalent data suitable for a BLP modelfor four categories. This did not apply to Dunnhumby's data, which however did notcover sufficiently representative product availability in the market for white goods.23 I am thankful to an anonymous referee for suggesting this point.
Table 8Parameter estimation of the AIDS for light bulbs.
N = 104. Estimation: seemingly unrelated regressions. Breusch–Pagan test of independence: Chi2(10) = 272.261***. Significance is as follows: * = 0.10; ** = 0.05; *** = 0.01.a Bootstrapped standard errors (1000 replications).b Elasticity of CFL is just below one (i.e., around 0.99999), while that of GLS is just above one (i.e., around 1.000001).
61L.A. Panzone / Energy Economics 38 (2013) 51–63
benefits of investments. Nevertheless, rising energy prices seem tonudge consumers into investing in energy efficiency, overcoming a statusquo bias and a product-specific endowment effect (see e.g. Kahnemanet al., 1991) that causes the postponement of EUP replacements.
A third important finding is that all efficient appliances (exceptA+ washing machines) are perceived as necessities. Therefore, con-sumers might value the environmental performance of EUPs lessthan other product attributes with private benefits from consumption(e.g. TV screen, refrigerator size). If this result was confirmed in futureresearch, the energy-efficiency gap could be reduced by bundling ahigh energy rating with other luxury attributes, increasing the per-ceived value of efficient EUPs. From a policy perspective, interven-tions releasing a disposable income constraint directly would be anadequate measure to reduce the energy efficiency gap only for A+washing machines.
A fourth and final result appears observing the impact of real ex-penditures and credit on the purchase of efficient EUPs. As expected,consumers seem to see investment in energy efficiency as costly, par-ticularly for goods with a short life like washing machines and televi-sions (Young, 2008). While an increase in real expenditures favoursthe purchase of efficient washing machines, it promotes inefficientrefrigerators and TVs. Periods where credit is available to consumersare instead characterised by high demand for A-rated washing ma-chines and A+ refrigerators only, although results do not establishcausality. Therefore, periodswith credit available for the purchase of ex-pensive complements of appliances (e.g. houses) are characterised bythe purchase of products that accommodate the need for a compromisebetween cost and environmental performance.
A limitation of this study, mentioned earlier in the article, is themissing availability of product characteristics, which inevitably influ-ence the choice of EUPs. In fact, consumers choose by trading off attri-butes and results indicate that energy-efficiency in some EUP mightgenerate less utility compared to other characteristics. The inability tofully account for product heterogeneity is a known limitation of theAIDS model (Nevo, 2000), and different modelling strategies could de-liver a more detailed picture of the fourmarkets under study. However,a superior performance in the estimation requires more expensivedatasets.22 Similarly, the coverage of the data (7%–37% of the UK mar-ket) might limit the generalisability of these results.23 While coveringan extensive amount of transactions, the samplemight be only partiallycomparable to national patterns of consumption. Furthermore, con-sumer behaviour could differ across store format and location, but thisdataset cannot capture a higher level of detail. Nevertheless, the datashould be comprehensive enough to present sensible and reliable re-vealed preference estimates.
A final limitation of this study is the scarce attention to the role ofthe discount rate in the energy efficiency gap. Consumers often re-frain from investing in efficient EUPs because of the uncertaintyover future energy and technology costs. The prospect of an imperfect
Table 9Elasticities of demand of residential electricity prices.
Significance is as follows: * = 0.10; ** = 0.05; *** = 0.01.a Bootstrapped standard errors (1000 replications).
62 L.A. Panzone / Energy Economics 38 (2013) 51–63
irreversible investment reduces the present value of net savings andincreases the implicit rate of discount (Jaffe et al., 2004). Consumersmight then wait for the investment anticipating the purchase of a bet-ter technology at the same price in the near future, given subjectiveexpectations over future energy prices. Because the data of this articleobserve equilibrium points where investment decisions are made, re-sults are estimated given the implicit discount rate of the shopper.Results cannot explain whether the observed efficiency gap is causedby high (private or social) discount rates or just an imperfect cost–benefit analysis (see Jaffe et al., 2004). Finally, future energy priceelasticities are likely showing a lower variance than reality becausethese prices are observed with certainty.
To conclude, this research draws a snapshot of the UKmarket for EUPsthat can be used as a reference point for future research. Few other anal-yses have reached the level of detail presented in this article, providingelasticities by energy-efficiency class for four different markets. The onlycomparable findings are those from Galarraga et al. (2011a), who findthat energy efficient washing machines are more price elastic than theirinefficient alternatives. However, they explore the demand for energy ef-ficiency and its premium, whereas this article focuses specifically on thedemand for energy efficient appliances for a variety of products. Impor-tantly, both this and previous studies support the importance of under-standing market barriers to energy efficiency to reduce the energyefficiency gap, and all results presented here suggests thatmore attentionshould be given to the importance of behavioural adjustment, consump-tion practices, and quality perceptions on consumer choices.
Acknowledgements
I am indebted to Ali Erbilgic and Mavourneen Pieterse (GHK) forexcellent research assistance, and to Deepak Kumar and ChrisGartside (Dunnhumby) for technical support. Comments from ChrisFoster, Chris Gartside, Lester Hunt, Andrew Jarvis, Juan Pablo Montero,Dale Southerton, Harry Tsang, Alistair Ulph, Evan Williams, and AdaWossink, as well as participants to the 19th Annual EAERE conferencehave been extremely helpful and appreciated. This project was entirelyfunded by DEFRA (grant number EV0709), and awarded to GHKConsulting (http://www.ghkint.com/). I am grateful to them for theirpermission to use the data for the present article. Finally, the sugges-tions of the editor (Richard Tol) and two anonymous referees havegreatly improved this work. All remaining errors are the author's soleresponsibility.
Supplementary data to this article can be found online at http://dx.doi.org/10.1016/j.eneco.2013.03.002.
References
Allcott, H., 2011. Social norms and energy conservation. J. Public Econ. 95 (9–10), 1082–1095.Allcott, H., Greenstone, M., 2012. Is there an energy efficiency gap? National Bureau of
Economic Research Working Article 17766. available online at http://www.nber.org/articles/w17766.
Allcott, H., Mullainathan, S., 2010. Behavior and energy policy. Science 327, 1204–1205.Anderson, C.D., Claxton, J.D., 1982. Barriers to consumer choice of energy efficient
products. J. Consum. Res. 9 (2), 163–170.Blanciforti, L., Green, R., 1983. An almost ideal demand system incorporating habits: an
analysis of expenditures on food and aggregate commodity groups. Rev. Econ. Stat.65 (3), 511–515.
Blundell, R., Robin, J.-M., 2000. Latent separability: grouping goods without weak sep-arability. Econometrica 68 (1), 53–84.
Brenčič, V., Young, D., 2009. Time-saving innovations, time allocation, and energy use:evidence from Canadian households. Ecol. Econ. 68 (11), 2859–2867.
Brennan, T., 2011. Energy efficiencypolicy: surveying thepuzzles. Resources for the FutureDiscussion Article No. 11-27. (Available at SSRN: http://ssrn.com/abstract=1977672or http://dx.doi.org/10.2139/ssrn.1977672).
Carman, H.F., 1972. Improving sales forecasts for appliances. J. Mark. Res. 9 (2),214–218.
Chalfant, J.A., 1987. A globally flexible, almost ideal demand system. J. Bus. Econ. Stat. 5(2), 233–242.
Dale, L., Fujita, K.S., 2008. An Analysis of the Price Elasticity of Demand for HouseholdAppliances. Energy Analysis Department, University of California Berkeley (availableonline at http://www.osti.gov/bridge/servlets/purl/929429-Vm1S9X/929429.pdf).
Dale, L., Antinori, C., McNeil, M., McMahon, J.E., Fujita, K.S., 2009. Retrospective evalu-ation of appliance price trends. Energy Policy 37 (2), 597–605.
Davis, L.W., 2008. Durable goods and residential demand for energy and water:evidence from a field trial. RAND J. Econ. 39 (2), 530–546.
Davis, L.W., 2011. The design and implementation of U.S. climate policy: evaluating theslow adoption of energy efficient investments: are renters less likely to have ener-gy efficient appliances? NBER Book Chapter. (Available at SSRN: http://ssrn.com/abstract=1907060).
Deaton, A., Muellbauer, J., 1980. An almost ideal demand system. Am. Econ. Rev. 70 (3),312–326.
DECC, 2011. Energy consumption in the United Kingdom: 2011. Available online athttp://www.decc.gov.uk/assets/decc/11/stats/publications/energy-consumption/2323-domestic-energy-consumption-factsheet.pdf.
Dhar, T., Chavas, J.-P., Gould, B.W., 2003. An empirical assessment of endogeneity issuesin demand analysis for differentiated products. Am. J. Agric. Econ. 85 (3), 605–617.
Dietz, T., Gardner, G.T., Gilligan, J., Stern, P.C., Vandenbergh, M.P., 2009. Household ac-tions can provide a behavioral wedge to rapidly reduce US carbon emissions. PNAS106 (44), 18452–18456.
Duarte, C.H., Schaeffer, R., 2010. Economic impacts of power electronics on electricitydistribution systems. Energy 35 (10), 4010–4015.
Dubin, J.A., McFadden, D.L., 1984. An econometric analysis of residential electric appli-ance holdings and consumption. Econometrica 52 (2), 345–362.
Edelstein, P., Kilian, L., 2009. How sensitive are consumer expenditures to retail energyprices? J. Monet. Econ. 56 (6), 766–779.
EU, 1998. Commission directive 98/11/EC of 27 January 1998 implementing CouncilDirective 92/75/EEC with regard to energy labelling of household lamps.
EU, 2010a. Commission Delegated Regulation (EU) No 1060/2010 of 28 September2010 supplementing Directive 2010/30/EU of the European Parliament and ofthe Council with regard to energy labelling of household refrigerating appliances.
EU, 2010b. Commission Delegated Regulation (EU) No 1061/2010 of 28 September2010 supplementing Directive 2010/30/EU of the European Parliament and ofthe Council with regard to energy labelling of household washing machines.
EU, 2010c. Commission Delegated Regulation (EU) No 1062/2010 of 28 September2010 supplementing Directive 2010/30/EU of the European Parliament and ofthe Council with regard to energy labelling of televisions.
Fasolo, B., McClelland, G.H., Todd, P.M., 2007. Escaping the tyranny of choice: whenfewer attributes make choice easier. Mark. Theory 7, 13–26.
Filippini, M., 1995. Electricity demand by time of use: an application of the householdAIDS model. Energy Econ. 17 (3), 197–204.
Fischer, C., 2008. Emissions pricing, spillovers, and public investment in environmen-tally friendly technologies. Energy Econ. 30 (2), 487–502.
Galarraga, I., González-Eguino, M., Markandya, A., 2011a. Willingness to pay and priceelasticities of demand for energy-efficient appliances: combining the hedonic ap-proach and demand systems. Energy Econ. 33 (1), S66–S74.
Galarraga, I., Heres, D., González-Eguino, M., 2011b. Price premium for high-efficiencyrefrigerators and calculation of price-elasticities for close-substitutes: combiningHedonic pricing and demand systems. J. Clean. Prod. 19, 2075–2081.
Gillingham, K., Newell, R.G., Palmer, K., 2009. Energy efficiency economics and policy.Resources for the Future, Discussion Article 09-13.
Green, R., Alston, J.M., 1990. Elasticities inAIDSmodels. Am. J. Agric. Econ. 72 (2), 442–445.Greene, D.L., 2011. Uncertainty, loss aversion, and markets for energy efficiency. Energy
Econ. 33 (4), 608–616.Hamilton, J.D., 2003. What is an oil shock? J. Econ. 113 (2), 363–398.Hausman, J.A., 1979. Individual discount rates and the purchase and utilization of
energy-using durables. Bell J. Econ. 10 (1), 33–54.Herrnstein, R.J., Loewenstein, G.F., Prelec, D., Vaughan, W., 1993. Utility maximization
and melioration: Internalitites in individual choice. J. Behav. Decis. Mak. 6 (3),149–185.
Jaffe, A.B., Stavins, R.N., 1994. The energy-efficiency gap. What does it mean? EnergyPolicy 22 (10), 804–810.
Jaffe, A.B., Newell, R.G., Stavins, R.N., 2004. Economics of energy efficiency. Encycl.Energy 2, 79–90.
Kahneman, D., 2003. Maps of bounded rationality: psychology for behavioral econom-ics. Am. Econ. Rev. 93 (5), 1449–1475.
Kahneman, D., Knetsch, J.L., Thaler, R.H., 1991. Anomalies: the endowment effect, lossaversion, and status quo bias. J. Econ. Perspect. 5 (1), 193–206.
Kilian, L., 2008. The economic effects of energy price shocks. J. Econ. Lit. 46 (4),871–909.
Larivière, E., Larue, B., Chalfant, J., 2000. Modeling the demand for alcoholic beveragesand advertising specifications. Agric. Econ. 22 (2), 147–162.
Linares, P., Labandeira, X., 2010. Energy efficiency: economics and policy. J. Econ. Surv.24 (3), 573–592.
Markandya, A., Arigoni Ortiz, R., Mudgal, S., Tinetti, B., 2009. Analysis of tax incentivesfor energy-efficient durables in the EU. Energy Policy 37 (12), 5662–5674.
Meier, A.K., Whittier, J., 1983. Consumer discount rates implied by purchases of energy-efficient refrigerators. Energy 8 (12), 957–962.
Metcalf, G.E., Weisbach, D., 2009. The design of a carbon tax. Harv. Environ. Law Rev. 33(2), 499–506.
Mick, D.G., Fournier, S., 1998. Paradoxes of technology: consumer cognizance, emotions,and coping strategies. J. Consum. Res. 25 (2), 123–143.
Mills, B.F., Schleich, J., 2010. What's driving energy efficient appliance label awarenessand purchase propensity? Energy Policy 38 (2), 814–825.
Mintel, 2010. Lighting — UK. report prepared by Mintel. available online at http://academic.mintel.com/sinatra/oxygen_academic/search_results/show&/display/id=479937/display/id=551737#hit1.
Mintel, 2011a. Fridges and freezers— UK. report prepared by Mintel. available online athttp://academic.mintel.com/sinatra/oxygen_academic/display/id=545551.
Mintel, 2011b. Televisions — UK. report prepared by Mintel. available online at http://academic.mintel.com/sinatra/oxygen_academic/search_results/show&/display/id=545447/display/id=594611#hit1.
Mintel, 2011c. Washers and dryers— UK. report prepared by Mintel. available online athttp://academic.mintel.com/sinatra/oxygen_academic/display/id=545551/display/id=545552/display/id=585228/display/id=585214#figure1.
Nevo, A., 2000. A practitioner's guide to estimation of random-coefficients logit modelsof demand. J. Econ. Manag. Strateg. 9 (4), 513–548.
Newell, R.G., Jaffe, A.B., Stavins, R.N., 1999. The induced innovation hypothesis andenergy-saving technological change. Q. J. Econ. 114 (3), 941–975.
Perino, G., Panzone, L., Swanson, T. (forthcoming). Motivation crowding in real con-sumption decisions: who is messing with my groceries? Economic Inquiry.
Poi, B.P., 2002. From the help desk: demand system estimation. Stata J. 2 (4), 403–410.Popp, D., 2002. Induced innovation and energy prices. Am. Econ. Rev. 92 (1), 160–180.Reiss, P.C., White, M.W., 2005. Household electricity demand, revisited. Rev. Econ. Stud.
72 (3), 853–883.Reiss, P.C., White, M.W., 2008. What changes energy consumption? Prices and public
pressures. RAND J. Econ. 39 (3), 636–663.Revelt, D., Train, K., 1998. Mixed logit with repeated choices: households' choices of
appliance efficiency level. Rev. Econ. Stat. 80 (4), 647–657.Richardson, I., Thomson, M., Infield, D., Delahunty, A., 2009. Domestic lighting: a high-
resolution energy demand model. Energy Build. 41 (7), 781–789.Sammer, K., Wüstenhagen, R., 2006. The influence of eco-labelling on consumer
behaviour — results of a discrete choice analysis for washing machines. Bus.Strateg. Environ. 15 (3), 185–199.
Shen, J., Saijo, T., 2009. Does an energy efficiency label alter consumers' purchasingdecision? A latent class approach based on a stated choice experiment in Shanghai.J. Environ. Manag. 90 (11).
Stokes, M., Rylatt, M., Lomas, K., 2004. A simple model of domestic lighting demand.Energy and Buildings 36 (2), 103–116.
Sutherland, R.J., 1996. The economics of energy conservation policy. Energy Policy 24(4), 361–370.
Vandenbergh, M.P., Steinemann, A.C., 2007. The carbon-neutral individual. N. Y. Univ.Law Rev. 82, 1673–1745.
Verbeke, W., Ward, R.W., 2001. A fresh meat almost ideal demand system incorporat-ing negative TV press and advertising impact. Agric. Econ. 25 (2–3), 359–374.
Warde, A., 2005. Consumption and theories of practice. J. Consum. Cult. 5 (2), 131–153.Young, D., 2008. When do energy-efficient appliances generate energy savings? Some
