Technology invention and adoption in residential energy consumption

    Technology invention and adoption in residential energy consumption. A stochastic frontier approach Giovanni Marin PII: DOI: Reference:

S0140-9883(17)30199-8 doi:10.1016/j.eneco.2017.06.005 ENEECO 3669

To appear in:

Energy Economics

Received date: Revised date: Accepted date:

21 September 2015 6 June 2017 14 June 2017

Please cite this article as: Marin, Giovanni, Technology invention and adoption in residential energy consumption. A stochastic frontier approach, Energy Economics (2017), doi:10.1016/j.eneco.2017.06.005

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT

RI P

T

Technology invention and adoption in residential energy consumption. A stochastic frontier approach Giovanni Marin1 Alessandro Palma2 1

SC

Department of Economics, Society and Politics, University of Urbino 'Carlo Bo', Urbino, Italy; SEEDS Sustainability Environmental Economics and Dynamics Studies, Ferrara, Italy. E-mail: [email protected] 2

AC

CE

PT

ED

MA NU

Corresponding author. IEFE, Bocconi University. Roentgen G. 1, 20136, Milan, Italy; CEIS, Faculty of Economics - University of Rome ‘Tor Vergata’, Via Columbia, 2, 00133 Rome, Italy; SEEDS Sustainability Environmental Economics and Dynamics Studies, Ferrara, Italy. E-mail: [email protected]

1

ACCEPTED MANUSCRIPT

SC

Abstract

RI P

T

Technology invention and adoption in residential energy consumption. A stochastic frontier approach.

CE

PT

ED

MA NU

In this paper we analyse the electricity consumption of a set of four traditional ‘white goods’ in a panel of ten EU countries observed over the period 1995-2013 with the aim of disentangling the amount of technical efficiency from overall energy saving using a stochastic frontier approach. The efficiency trend is modelled as a function of energy efficiency policies and innovation dynamics that combines invention and adoption processes using specific patents weighted by granular production data and worldwide bilateral import flows. Our model also accounts for potential endogeneity arising when innovation processes and economic growth are considered. With this replicable approach, the stochastic frontier framework allows for explicit modelling of innovation processes. Our results show that the efficiency component is related to changes in the energy efficient technological content of appliances. The 'international' component represents a predominant share of technological advancement and exerts a significant influence on the transient efficiency. Our evidence calls for an active role to be played by policy makers in focusing on innovation and trade policies in order to achieve more ambitious energy efficiency targets.

AC

JEL: C23, C26, O33, Q55, Q41. Keywords: energy efficiency, energy policies, technological adoption, electrical appliances, stochastic frontier analysis, residential sector.

2

ACCEPTED MANUSCRIPT

1 Introduction

AC

CE

PT

ED

MA NU

SC

RI P

T

Energy efficiency (EE) constitutes one of the most cost-effective strategies for reducing the amount of primary energy consumption, thus contributing to increased energy security and lower greenhouses gas emissions (GHGs). It is thus not surprising that EE has scaled up its role in EU climate and energy strategies. The "2030 Climate and Energy Framework" approved by the European Commission in 2014 combines a target for GHGs reduction (40% by 2030 compared with 1990 levels) with a specific target on energy efficiency in the order of 27% of energy efficiency improvements by 2030 with respect to the business as usual scenario (EC, 2014). Despite numerous studies that analyse the contribution that energy efficiency makes to reducing energy consumption, we identify two major gaps in the literature. First, although EE is intrinsically related to technology (Linares and Labandeira, 2010; Hartman, 1979; van den Bergh, 2011), the role of innovation is not explicitly accounted for in existing studies. The second gap relates to the limited focus of the existing literature on the household sector, which nowadays represents a major concern for policy makers given its increasing role in the portfolio of energy services. In this paper we attempt to fill these gaps by employing an original panel dataset of 10 EU countries observed over the period 1995-2013 to analyse the drivers of households’ electricity demand for a set of four traditional large electrical appliances. We go beyond the original contribution of Filippini et al. (2014), who analysed the role of policies as a major determinant of efficiency gains, by enriching the framework with explicit modelling of EE-related innovation (invention and adoption). In particular, we account for innovation dynamics to explain the mechanism that triggers a substantial reduction in electricity consumption for major large home appliances. A stochastic frontier analysis (SFA) is employed to disentangle the amount of energy saving observed in the demand estimation due to improved technical efficiency. The derived demand frontier allows for explicit modelling of technical efficiency which includes the contribution made by both EE policies and technology invention and adoption as sources of efficiency improvements that impact overall electricity consumption. In line with previous studies, our results confirm that the gains in technical efficiency and associated consumption reduction are related to the number of in-force EE policies which include demand-pull instruments aimed at stimulating households to purchase more efficient appliances (e.g. labelling of appliances, price schemes, etc). However, this evidence does not account for the role of supply-push regulations (e.g. incentives to manufacturers for eco-design) and other endogenous firms’ activities (e.g. voluntary agreements based on corporate social responsibility and market opportunities for better technologies) which guarantee a growing availability of innovative energy efficient appliances on the market. Both these components are captured by our innovation proxy. When the latter enters the efficiency equation, the significance of policy disappears in favour of a strong significant impact of our innovation proxy. The magnitude of the estimated impact of our indicator of technological change, which also accounts for novel technologies embodied in imported electrical appliances, induces a consideration to be made regarding the importance of looking at the key complementary roles of supply side regulations and trade policies, whose effects trigger the invention process of more advanced appliances and its diffusion in (international) markets. The rest of the paper proceeds as follows. In Section 2, we describe the conceptual framework on which our analysis is based and review some relevant studies that analyse

3

ACCEPTED MANUSCRIPT

RI P

T

the determinants of EE in the residential sector. We stress the limitations of these studies and introduce the main contributions to our analysis. In Section 3, we describe the data and empirical strategy used to estimate the electricity stochastic frontier demand function. Section 4 presents the econometric results together with the efficiency scores obtained from our model and Section 5 concludes the paper with some policy implications. The limitations of the present work and further research insights are also discussed.

2 Conceptual framework

SC

2.1 Trends of residential electricity consumption and EE

AC

CE

PT

ED

MA NU

Official statistics (EC, 2012a; Bertoldi et al. 2016) suggest that the residential sector accounted for about 30% of total final electricity consumption in the EU27 (year 2010) and that such a share does not seem to be slowing down. In particular, traditional large home appliances (freezers, refrigerators, washing machines and dishwashers) were responsible in 2007 for 25% of households’ electricity consumption as opposed to other appliances such as information and communication technologies characterized by smaller energy needs than the so-called ''white appliances’’ (Bertoldi et al. 2016)1. Moreover, home appliances generally consume electricity instead of renewable fuels or direct combustion fuels, with a significant carbon footprint in countries where electricity production is carbon intensive (Cabeza et al., 2014). According to IEA forecasts, appliance consumption is expected to increase at a higher rate than the building sector as a whole (IEA, 2013) given the increasing demand for new goods. Since the latter are crucial to fulfilling primary needs such as food conservation or washing, they can be characterized as fully complementary goods to household dwellings. It is thus not surprisingly that white appliances, in particular refrigerators, have almost reached saturation point among EU dwellings (IEA, 2009) and that their market is characterized by a high rate of substitution of old equipment rather than by an increase in household stock. Consequently, the residential sector represents a special target of the EU mitigation strategies to ensure future patterns of decreasing electricity consumption and a low-carbon transition (EC, 2011; IEA, 2013; OECD, 2003). Among others, EE emerges as an effective, low-cost strategy to reduce households' electricity consumption without limiting the utility associated with energy services (Ramos et al., 2015). EE is strongly linked to the technological level of the equipment that is used to obtain a certain energy service. For instance, a more efficient dishwasher provides a higher number of washes while using the same or less electricity. The availability of new energy efficient appliances developed by manufacturers and progressively adopted by households to replace older ones represents a key driver of electricity saving. Even though constituted by mature technologies, both cooling and washing appliances still offer a great saving potential given their widespread presence among a multitude of household dwellings. A study by McKinsey (2009) points to the cost-effectiveness of EE gains deriving from electrical appliances compared with those deriving from other sectors and a more recent one (McKinsey, 2012) shows how the impacts of technology, 1

The portfolio of energy services available for households has massively increased over the last 40 years with a strong penetration of new devices and appliances aimed at satisfying these services. See Burwell and Swezey (1990).

4

ACCEPTED MANUSCRIPT policy regulations and consumer behaviour are likely to transform the EU residential energy market in the coming years. 2.2 Barriers to EE and the role of policies

AC

CE

PT

ED

MA NU

SC

RI P

T

Despite the fact that EE gains imply a large saving potential, market forces alone do not seem to achieve this potential. Several studies point out that available EE technologies are often adopted at sub-optimal levels, identifying barriers of a different nature (Brown, 2004; Jaffe et al., 2004). This phenomenon is known as the “EE gap” and can be defined as the perceived gap in uptake of existing energy efficient technologies despite the fact that the latter are characterized by positive net present values (Jaffe and Stavins, 1994; Gillingham and Palmer, 2014). This translates into slower-than-optimum paces of EE technology adoption (demand side) and, consequently, in weaker market stimuli for firms to innovate (supply side). The studies also stress the active role of regulation in boosting technological development and adoption which facilitates the replacement of older equipment with more efficient models. Consequently, the initial political interest in increasing residential EE is nowadays a political priority considering the multiple benefits of EE. This is translated into a growing, complex regulatory activity aimed at closing the EE gap with a combination of EU legislation (energy labelling and minimum energy performance standards2), national programmes (e.g. purchase incentives in Italy, price rebates in Spain, supplier obligations and White Certificate schemes in France, Italy and the UK) and Corporate Social Responsibility strategies (e.g. voluntary agreements of manufactures3). In particular, the ‘Eco-design Directive’ for Energy-Using Products (EuP Directive 2005/32/EC), the introduction of energy labelling for electric devices (Directive 92/75/ECC) or more recently, the Energy Efficiency Directive approved in 2012 (EC, 2012b), establish a set of binding measures to help the EU reach ambitious energy efficiency targets. Several empirical studies have analysed the effectiveness of policies aimed at improving EE in the residential sector. Jaffe and Stavins (1995) measure the impact of energy prices, adoption subsidies and building codes on the home EE level in the United States between 1979 and 1988, and find that government subsidies have a stronger effect on the buildings’ average level of EE than that induced by increasing energy prices. Unlike technology standards, energy taxes were associated with an increasing adoption of new technologies. Even though the effect is found to be relatively small, the authors conclude that building codes are often set too low to be effective. Newell et al. (1999) test the hypothesis of policy-augmented price-induced innovation that relies on sale data of room and central air conditioners as well as gas water heaters in the 19581993 period. They find positive relations between EE performance and the technology turnover. The latter is led both by increasing energy prices or lower appliances' prices. They also find a positive and significant association between regulatory activity 2

The Eco-design of the Energy-Using Products Framework Directive 32/2005/EC (Eco-design Directive), the end-use energy efficiency and energy services Directive 32/2006/EC (ESD), the Energy Performance of Buildings Directive 91/2002/EC (EPBD, under recast) as well as the Labelling Directive 75/1992/EC (under recast) contribute significantly to implementing the energy-saving potential in the European Union. Accompanying and completing EU legislation, many energy efficiency measures concerning financial incentives, supplier obligations, information, etc. have also been adopted by the Member States. 3 An example is the Conseil Européen de la construction d’appareils domestiques (CEDEC).

5

ACCEPTED MANUSCRIPT

AC

CE

PT

ED

MA NU

SC

RI P

T

represented by energy labelling requirements and technological improvement. Buck and Young (2007) use cross-sectional data on energy use to derive efficiency scores for different types of commercial buildings in Canada. Accordingly, they find a relatively high level of efficient and significant differences between government-owned buildings and those owned by non-profit organizations. The authors acknowledge that, due to data limitations, the effect of new technology adoption is not fully captured by their model. Filippini and Hunt (2012) use a balanced panel deriving from the US-EIA database to analyse residential energy consumption in 48 US States over the 1995-2007 period. They find an inconsistent pattern across US States between standard energy intensity indicators and energy efficiency scores deriving from the stochastic frontier approach, suggesting the need for further investigation in this direction. More recently, Filippini et al. (2014) focus on the impact of government policies that aim to improve energy efficiency in the residential sector. Although a large number of in-force policy instruments exist in the EU, they find room for efficiency gains and large variability across countries. Building on the limited effectiveness of traditional EE policies which rely on perfect information and rational-decision making, Ramos et al. (2015) point to the key role of a well-designed instruments mix which must be conceived on the basis of specific barriers affecting each sector. Similar conclusions are stressed by Bigano et al. (2010), who analyse the implications of regulations affecting the energy efficiency performances on a panel of 15 EU countries and Norway in several energy sectors. With no exceptions, all these contributions show that regulation activity in all the EU countries cannot be disregarded in an analysis of household consumption. Together with the barriers discussed above, electricity saving via efficiency gains can be partly or completely offset by the rebound effect (Khazzoom, 1980; Sorrell and Dimitropoulos, 2008). Ceteris paribus, gains in efficiency increase the amount of disposable income to be allocated to energy spending. Consumers may thus increase the energy demand for the same energy service (direct rebound effect) or for other energy and non-energy goods (indirect rebound effect). While the impact of direct rebound could be evaluated by considering the price elasticity and income elasticity for the demand of a given energy service, the assessment of indirect rebound effect is more complex and requires a general equilibrium perspective (de Miguel et al., 2015). An important implication of the rebound effect is that it may severely limit the effectiveness of energy efficiency policies. At the same time, the literature analysing energy consumption may overestimate the impact on saving if both direct and indirect rebound effects are not considered4. Accounting for the significance of the rebound effect is therefore important if we want to properly evaluate the effectiveness of an energy policy instrument that aims to promote energy efficiency gains. 2.3 The role of technology In recent years, the growth of EE technologies has been impressive and documented in several empirical works (Costantini et al., 2015; Noally, 2012; Verdolini and Galeotti, 2011, among others). Even though these studies do not relate the innovation pattern to EE gains, they stress the key role of policy-induced innovation in speeding up energy saving and emissions abatement. Surprisingly, we observe very few contributions in the 4

For instance, Chitnis and Sorrell (2015) provide a calculation method for both direct and indirect rebound, together with an empirical application which shows that if indirect rebound is not included in the analysis, the amount of total rebound can be underestimated.

6

ACCEPTED MANUSCRIPT

AC

CE

PT

ED

MA NU

SC

RI P

T

existing literature that specifically focus on the relationship between innovation, efficiency performance and reduction in energy consumption. In our view, this constitutes a gap that we attempt to fill. In most empirical works using SFA, technology is often modelled as a latent variable with no explicit treatment and measurement of innovation (Filippini et al., 2014). Ideally, we would like to capture the contribution of the technology employed in the ‘production process’ of a given energy service to energy saving. This contribution potentially translates into a large impact on overall energy saving at the aggregate level if multiplied by the number of appliances sold on the market and operating in households’ dwellings. Given the absence of reliable data on the total stock of devices that also account for their heterogeneous technological characteristics, it is often difficult to operationalize this modelling approach. Once a technology is invented and available on the market, its adoption rate, slow in the first phase, rapidly accelerates up to a saturation point in which the adoption speed of the new technology reaches its maximum and declines in favour of new technologies introduced into the market (Griliches, 1957; Geroski 2000; Diaz-Rainey and Ashton, 2015). In EE technologies, the typical S-shaped curve traced by the level of technology uptake has different explanations. Important demand factors can be identified in the adopters' propensity to purchase new equipment which in turn depends on the awareness of energy saving potential, the access to technical information, and the availability of financial means. The substantial heterogeneity in preferences among consumers leads to differences in the expected returns to adoption. However, these differences tend to be reduced over time as the cost of new technologies falls and information becomes increasingly available. Heterogeneity in the technology adoption rate changes according to the appliance that is considered (Jaffe et al., 2004; Fernandez, 2001) since the longer the expected lifetime of the appliance, the more the consumer faces long-term energy savings concerns, also considering the growing trend in energy prices that has occurred over the past decades (Popp, 2002) and recently in Europe (Bertoldi et al., 2016). Supply-side factors are also equally important since various stages are required to develop and commercialize technologically advanced appliances. These include the development of EE technologies through R&D activities, manufacturing, transportation, commercialization and recycling. Hence, the value chain of electrical appliances consists of a number of actors that synergistically work to increase the renovation rate of older equipment (Thomas, 2015). Following the conceptual contributions of scholars in the innovation studies field (Stoneman, 1983; 2001, among others), three main stages in the innovation process can be identified, namely invention (i.e. the generation of new ideas), innovation (i.e. the development of new ideas into marketable products and processes) and adoption (in which the new products and processes spread across the potential market). In order to understand how the consumption pattern changes as new technologies are introduced and employed, the stage of adoption assumes a crucial role (Karshenas and Stoneman, 1993). In our case, a new EE technology is initially developed and embodied in the appliance. Subsequently, the appliance is diffused in the market and made accessible to consumers (both in terms of logistics as well as economic affordability). Finally, the appliance is purchased (i.e. adopted) by consumers and used. In this respect, data and metrics for measuring both manufacturers' innovative performances as well as the level of technology adoption are particularly important. Data on specific product characteristics would represent a good source for analysing the technological level of

7

ACCEPTED MANUSCRIPT

PT

ED

MA NU

SC

RI P

T

appliances, but they are difficult to collect for long time series. On the other hand, technology-input measures, such as private R&D expenditures, are often not publicly available and do not provide information on the rate of success of research efforts. Different solutions have been adopted to overcome these limitations. For instance, Datta and Gulati (2014) investigate the implications of consumers’ behaviour in response to EE gains that arise from the use of more efficient appliances by employing energy labels and codes as a measure of efficiency performance. Nevertheless, this approach provides a poor representation of the technology portfolio embodied in the appliances under scrutiny, with a raw distinction among the different technology advances implemented by the multitude of appliance manufacturers. In the SFA framework, it has been argued that technology's contribution can be indirectly captured by a number of factors such as, for instance, price and income effects, the energy demand trend and climate variables (see Filippini and Hunt, 2012). In this respect, Filippini et al. (2014) argue that one limitation of this approach lies in the difficulty of disentangling the amount of saving given by an efficient combination of inputs (energy, labour, capital) with old technologies or non-efficient use of inputs though households employ bestavailable technologies. These limitations are mainly due to the lack of data on consumers’ behaviour, heterogeneity in the level of electricity consumption and on the technologies employed. Given that the most important aim of EE policies for electrical appliances is to accelerate the replacement process of older equipment with newer ones through different mechanisms, a possible approach to capturing the impact of incremental adoption is to use the policies as a source of variation of technical efficiency, a method tested by Filippini et al. (2014). However, this approach cannot be used to separate the effects of technology on changing technical efficiency or the ones provided by incremental invention and the adoption of new equipment.

CE

3 Data and model specification

AC

Our analysis of the drivers of the demand for electricity of home electrical appliances considers a balanced panel of ten EU countries5 observed over the period 1995-2013. Among the different methodological approaches used to measure the level of technical efficiency and its determinants, we rely on the SFA framework, a parametric empirical technique which can be used to estimate the frontier level of theoretical and actual efficiency of a given production system in the well-known framework of the neoclassical production function (Aigner et al., 1977; Meeusen and van den Broeck, 1977)6. Specifically, we employ recent developments of the SFA technique which allows to incorporate exogenous influences on the level of efficiency by means of auxiliary variables (see equation 3). Although SFA is not exempt from drawbacks such as the imposition of a predetermined functional form and its reliance on strong assumptions regarding the distribution of both inefficiency and idiosyncratic error terms, it has been extensively exploited in the literature of energy economics (see Buck and Young, 2007; Boyd, 2008; Filippini and Hunt, 2012; Stern, 2012, among others). Several studies have successfully applied the SFA approach to the framework of households' energy demand. Accordingly, households are assumed not to demand electricity per se, but for the need of energy services which are satisfied by using 5

Austria, Denmark, France, Germany, Greece, Italy, Netherlands, Slovenia, Sweden, United Kingdom. Other possible approaches can be Data Envelopment Analysis (DEA) (Thore et al., 1994) and decomposition methods (Ang, 1995).

6

8

ACCEPTED MANUSCRIPT

SC

∗ 𝐸𝑖,𝑡 = 𝛼𝑖 + 𝑥 ′ 𝑖,𝑡 𝛽 + 𝜖𝑖,𝑡

RI P

T

different electrical appliances (Linares and Labandeira, 2010). In more detail, households purchase and combine inputs to gain utility (i.e. energy services) represented by a composite of energy commodities (Filippini, 1995; Filippini and Hunt, 2012; Filippini and Pachauri, 2004). In line with this definition, the (derived) energydemand stochastic frontier model (EDSFM) provides the minimum energy input (i.e. electricity consumption) used by a household for a given level of (unobservable) output (i.e. energy services)7. In a panel setting, the EDSFM is described by the following equation:

𝜖𝑖,𝑡 = 𝜈𝑖,𝑡 + 𝑢𝑖,𝑡

(1) (2)

AC

CE

PT

ED

MA NU

∗ where 𝐸𝑖,𝑡 represents the theoretical minimum consumption demand frontier expressed in kWh and demanded by households for using electrical appliances. 𝑥𝑖,𝑡 is the vector of inputs and controls that influence the demand for energy services which include end-use electricity prices per kWh, GDP per-capita income, both expressed in purchase-power parity (PPP) 2005 US dollars, average size of dwellings in squared meters, the urban population as a share of total population and a vector 𝛼𝑖 of country fixed effects. 𝛽 is the vector of parameters of interest to be estimated. Dependent and independent variables (with the exception of the share of urban population) are expressed in logarithms. The estimation results have to be interpreted as elasticity of electricity consumption with respect to the vector of covariates. Descriptive statistics and data sources for these variables are reported in Table A1. The error term 𝜖𝑖,𝑡 is constituted by a technical inefficiency component 𝑢𝑖,𝑡 , that follows a generic distribution ℱ𝑢 with support defined over ℝ+ to be interpreted as a measure of the technical inefficiency in using energy to provide a certain energy service, and an idiosyncratic component 𝜈𝑖,𝑡 , that includes measurement errors and is assumed to be normally distributed. Furthermore, by denoting 𝓏𝑖,𝑡 a vector of exogenous variables affecting the level of inefficiency and 𝜙 a vector of unknown parameters to be estimated, the statistical distribution of the inefficiency term can be explicitly modelled as follows: ′ 𝑢𝑖,𝑡 = 𝑧𝑖,𝑡 𝜙

(3)

According to the stochastic frontier framework, the actual demand level 𝐸𝑖,𝑡 equals the ∗ theoretical frontier 𝐸𝑖,𝑡 , plus the one-sided inefficiency error 𝑢𝑖,𝑡 , whose distribution ′ depends on the vector of auxiliary variables 𝑧𝑖,𝑡 which include, in our case, a policy indicator and a proxy for technology penetration. Income and electricity prices constitute the inputs of the (derived) electricity demand that we estimate for two groups of home electrical appliances, namely cooling appliances (refrigerators and freezers) and washing appliances (dishwashers and washing machines). Data on appliance-specific electricity consumption derive from the Odyssee database, developed by Enerdata in collaboration with several national energy

7

See also Filippini and Hunt (2015).

9

ACCEPTED MANUSCRIPT

AC

CE

PT

ED

MA NU

SC

RI P

T

agencies, under the supervision of the European Commission8. In addition, we employ a set of additional controls. The first is the average size of households' dwellings since the size of a house has been recognized in the literature as an important socio-economic determinant of residential energy consumption. Larger houses are likely to include more appliances or, at least, 'larger' appliances. The inclusion of this control allows us to account for possible size-effects in the energy demand (ETC-SCP, 2013; Kaza, 2010). Some studies based on individual data also point to a significant role of behavioural and socio-economic factors in determining electricity consumption via the adoption of EE appliances and energy-conscious use of electrical appliances (Gans et al., 2013; Kavousian et al., 2013; Schleich et al., 2013). However, behavioural data including information on energy consumption conditional to a level of awareness of environmental-related problems or instant consumption volumes are difficult to find for a large set of countries. We thus consider the simpler hypothesis that the households’ electricity demand varies according to the characteristics of the urban context. To this end, we include the share of urban population over the total population to control for the behaviour of different populations. As discussed in Section 2.2, a further issue that deserves attention is the potential presence of a rebound effect. Both theoretical and empirical contributions show that the impact of rebound is proportional to the values of price elasticity for the associated energy service. Non-linear behavioural response can also characterize the priceelasticity of some energy services, in particular the ones where the consumer’s responsiveness is more or less constant up to a saturation point (also called ‘comfort zone’) beyond which the rebound may grow more than proportionally. For instance, in the heating sector, the price-elasticity and associated rebound can vary substantially if measured before and after a given room temperature. This is because once an individual achieves a comfort environment, his marginal willingness to pay for additional heating service decreases in favour of other goods or services. In the SFA framework, Orea et al. (2015) point out that the standard EDSFM as proposed by Filippini and Hunt (2012) and largely employed by other authors implicitly imposes a zero rebound effect. This is in contrast with the large empirical literature that consistently shows positive rebound values in the residential sector (Greening et al., 2000; Saunders, 2013, among others). At the same time, Orea et al. (2015) point out that the rebound effect is likely to be moderate when the elasticity of energy demand with regard to changes in energy efficiency is low. Hence, we may expect a minimum presence of the rebound effect in the traditional appliances sector. By fulfilling primary needs such as preserving food or washing clothes, these goods are characterized by a low price-elasticity and saturation effect (Chakravarty et al., 2013; Davis, 2008; Ek and Soderholm, 2010; Mills and Schleich, 2010). Moreover, some of them such as refrigerators and freezers operate continuously, with a very limited possibility of changing their operation status and associated saving performance, at least in the short run. 3.1 EE policies

8

Electricity consumption by electrical appliances (kWh/appliance) is usually estimated using calculation procedures that are specific to each appliance type: for instance, for washing machines and dishwashers, it is calculated as the electricity consumption per cycle multiplied by a number of cycles per year. For refrigerators, it is calculated as the specific electricity consumption per litre multiplied by the average size of the stock in litres and multiplied by 365 days.

10

ACCEPTED MANUSCRIPT

PT

ED

MA NU

SC

RI P

T

Policies aimed at promoting EE in the residential sector face barriers of different nature, such as technical, economic, political, and social impediments. In light of this, it is often necessary the implementation of multiple policy instruments which allow for a joint and coordinated policy action (Sovacool, 2009). According to Costantini et al. (2017) and Del Rio et al. (2010), three main policy pillars can be distinguished: demand-pull instruments, which include market based and command and control instruments aiming to enlarge the size of the market demand for new technologies; technology-push instruments, which mainly include R&D activities aimed to increase the supply of new scientific and technological knowledge; soft and systemic instruments. Soft instruments represent a broad domain of information and voluntary policy tools aimed at enhancing the consumer awareness on potential EE benefits. Systemic instruments address potential problems that limit the transmission mechanisms of more specific policy tools or affect the functioning of the EE sector and system as a whole (Wieczorek and Hekkert, 2012). Most of these instruments are included in the IEA Energy Efficiency Policies and Measures Database, which collects systematic and comprehensive information on the implemented policies for, among other sectors, residential appliances (refrigeration, cooking and laundry) for all the countries in our sample. Following Nesta et al. (2014) and Filippini et al. (2014), we use the cumulated number of EE policies to account for the role of regulation activities in triggering energy efficiency. Our policy variable is built as the cumulated count of policies that varies over years and across countries, and includes a combination of different policy dummies. Though in a crude way, this modelling choice captures both the intensity and variety of policy stimulus. The policy index (in logarithm) represents our first auxiliary variable used to model the variance of technical efficiency.

CE

[Error! Reference source not found. about here]

AC

Error! Reference source not found. shows the growing number of policy instruments progressively implemented in our sample of countries. This pool of regulations is mainly aimed at incentivising the replacement of older appliances beyond the behavioural response of consumers to mere electricity savings. As previously discussed, the growth in number and type of policies adopted in our sample of countries has been remarkable, with the exception of Greece and Slovenia which show absence of policy regulations to promote EE in home appliances. 3.2 Innovation dynamics The second auxiliary variable for inefficiency modelling accounts for innovation dynamics. The methodology we propose has several advantages. First, it considers the innovation process as a whole since the invention and adoption of EE technologies are explicitly considered. Both these innovation components are modelled as continuous variables without discrete shifts to produce a more realistic representation of innovation processes. Moreover, by weighting the technology stock by both domestic production and import flows, we also capture the effects of demand-pull EE policies, trade barriers and other mediating factors of the appliances' value chain that contribute to the market's performance. Finally, our indicator is based on objective data and can be generalized to a multitude of goods or sectors. Our method includes two data sources. Patent data constitute our measure of

11

ACCEPTED MANUSCRIPT

MA NU

SC

RI P

T

technology invention. Despite some limitations, patents represent a widely-used data source in the economics of innovation (Hall et al., 2005; Jaffe and Trajtenberg, 2004; Lanjouw et al., 1998; Malerba and Orsenigo, 1996; Popp, 2006, among others) since they provide a wealth of information on the nature of the invention and the applicant for a rather long time series. Patent data frequently represent the direct result of R&D processes, a further step towards the final output of innovation, which is useful knowledge through which firms are able to generate new profits. The patent database adopted here employs the Y02 tagging scheme of Cooperative Patent Classification (CPC) based on patent classes for energy efficiency technologies for the residential sector, further specified for the sub-sector of electrical appliances.9 Table B1 in the Appendix describes the CPC classes that were relevant to our study. As a result, we collect a total of 559 unique patent applications filed at the European Patent Office (EPO) and belonging to the four appliances, namely freezers and refrigerators, washing machines and dishwashers. Our patent sample has been sorted by priority date and assigned to the applicant’s country10. In order to capture both past and recent innovative efforts, the patent stock Πi,t has been calculated as follows: ∞

Π𝑖,𝑡 = ∑ 𝜋𝑖,𝑡−𝑠 𝑒 −𝛽1 (𝑠) (1 − 𝑒 −𝛽2 (𝑠+1) )

(4)

ED

𝑠=0

CE

PT

where 𝜋𝑖,𝑡 indicates the patent count in year t for applicants based in country i, 𝛽1the rate of decay that captures the obsolescence of older patents and 𝛽2 the rate of adoption that accounts for the delay in the adoption of knowledge. Following Popp (2003), these parameters are set to 0.1 and 0.25 respectively. This modelling choice treats the technology stock as a cumulative process, but at the same time, accounts for the obsolescence effect, i.e., as new technologies are available, older patents become less profitable (Hall, 2007; Evenson, 2002).

AC

[Error! Reference source not found. about here]

Error! Reference source not found. reports the value of the patent stock in energy efficient technologies for electrical appliance for the ten top patenting countries worldwide in 1995 and 2013. The concentration of patenting in energy efficient technologies for electrical appliances was very high both in 1995 and 2013: about 95 percent of the patent stock was concentrated in the ten most important patenting 9

A possible limitation of this approach is that in the case of green technologies, standard international patent classifications only partially represent the whole range of sub-fields that characterize complex technological domains such as EE (Barbieri and Palma, 2017). 10 A possible limitation when patents are employed as a measure of innovation output is represented by the high heterogeneity in their value (Griliches, 1998, among others). It is thus necessary to control for patent quality. In this respect, it is worth noting that EPO applications are much more expensive than applications to national patent offices and inventors typically apply to the EPO if they have strong expectations regarding the invention's economic returns. As a consequence, by considering EPO patents instead of the ones filed at national patent offices provides a “quality hurdle which eliminates applications for low-value inventions” (Johnstone et al., 2010, p. 139). Moreover, the exclusive use of EPO patents maximizes the likelihood of capturing inventions that are expected to be commercialized in the EU market.

12

ACCEPTED MANUSCRIPT

SC

RI P

T

countries. It is interesting to note, however, that the relative ranking of countries changed substantially over the period 1995-2013. The technological leader in 1995 was the US, followed closely by Italy and Germany. The 2013 ranking features Germany as the country with the largest patent stock, followed by South Korea and Italy, while the former leader (US) only ranks fourth. Moreover, new emerging countries such as China and Turkey also gained importance. Finally, many countries, even EU members, do not patent but purchase more efficient electrical appliances from abroad. Given that patents represent only the first stage of an innovation process, we also derive a proxy of technological adoption by employing appliance-specific trade and production data which can be used to capture both the foreign and domestic penetration of EE electrical appliances sold in the market.

MA NU

[Table 1 about here]

AC

CE

PT

ED

Table 1 reports information on the geographical source of purchases of electrical appliances (dishwashers, washing machines, fridges and refrigerators) by domestic consumers11. For each country i, we calculate the ratio between the purchase of electrical appliances from country j as a share of overall purchases of electrical appliances by domestic consumers (total import of electrical appliances plus domestic production in i net of export of electrical appliances by country i). Italy and Germany were the most important partners for the foreign supply of electrical appliances for our sample of ten countries, both in 1995 and in 2013, with the only exception being Sweden (which was the main partner for Denmark in 2013). A reasonably important role was also played by France (which ranked second twice in 1995 and 2013). The domestic component was larger than 50 percent in 1995 for Italy and Slovenia whereas it only remained above 50 percent in Slovenia in 2013. Among non-EU countries, only Switzerland and the US were in the top 3 partners for, respectively, Austria (1995, third position, accounting for about 3 percent of purchases) and the Netherlands (1995, third position, accounting for about 5 percent of purchases). Our measure of innovation creation and adoption of new energy efficient technologies combines information on the invention of new energy efficient technologies for electrical appliances (worldwide) and the diffusion of these technologies through trade. Our indicator of technology includes both a domestic and a foreign component (𝑇i,t𝐷𝑜𝑚 and 𝑇i,t𝐹𝑜𝑟 , respectively). We assume that the technologies embodied in electrical appliances enter the national markets through domestic appliances production and 11

We selected the following CN8 (Combined Nomenclature 8-digit) codes for measuring bilateral trade flows in the COMEXT database: 8418 "Refrigerators, freezers and other refrigerating or freezing equipment, electric or other; heat pumps; parts thereof (excl. air conditioning machines of heading 8415)", 8422 "Dishwashing machines; machinery for cleaning or drying bottles or other containers; machinery for filling, closing, sealing or labelling bottles, cans, boxes, bags or other containers; machinery for capsuling bottles, jars, tubes and similar containers; other packing or wrapping machinery, incl. heat-shrink wrapping machinery; machinery for aerating beverages; parts thereof" and 8450 "Household or laundry-type washing machines, incl. machines which both wash and dry; parts thereof". We selected the following NACE (rev. 1.1) codes for measuring domestic production and total import and export of appliances: 29711110 "Combined refrigerators-freezers; with separate external doors", 29711133 "Household type refrigerators", 29711135 "Built-in refrigerators", 29711150 "Freezers of the chest type; capacity =< 800 litres", 29711170 "Freezers of the upright type; capacity =< 900 litres", 29711200 "Dishwashers", 29711330 "Fully-automatic washing machines; capacity =< 10 kg", 29711350 "Non-automatic washing machines; capacity =< 10 kg".

13

ACCEPTED MANUSCRIPT

3 𝐷𝑜𝑚 𝑇𝑖,𝑡

MA NU

SC

RI P

T

foreign import flows, the latter expanding the internal supply of energy efficient appliances. Hence, in this stage, innovative EE appliances are sold by firms to households. In order to capture the impact of technology embodied in the appliances and actually sold in the market, each national patent stock is multiplied by the value of domestically-manufactured appliances. However, a relevant share of technical efficiency also derives from appliances purchased abroad and used by households in the national territory. As suggested by Shih and Chang (2009) and more recently by Costantini and Liberati (2014), well-established international market relationships represent a good means for testing the degree of embodied technology adoption. Accordingly, patent stocks belonging to foreign manufacturers are weighted by the corresponding partner-specific bilateral import flow. The values of both the domestic and foreign technology stocks have been divided by the total monetary value of appliances purchased by final consumers in a given country, calculated as the domestic production less the export share to which we summed the imported production from foreign countries. In a formula, the two technology stocks are calculated as follows: = ∑( 𝑘=1

ED

3

Π𝑖𝑘 (𝑌𝑖,𝑡 − 𝐸𝑥𝑝𝑖,𝑡 ) ) 𝑌𝑖,𝑡 − 𝐸𝑥𝑝𝑖,𝑡 + ∑𝑖 𝐼𝑚𝑝𝑖,𝑡

𝐹𝑜𝑟 𝑇𝑖,𝑡

= ∑(

PT

𝑘=1

k ∑𝑗≠𝑖 Πj,t 𝐼𝑚𝑝𝑖,𝑗,𝑡

𝑌𝑖,𝑡 − 𝐸𝑥𝑝𝑖,𝑡 + ∑𝑖 𝐼𝑚𝑝𝑖,𝑡

𝑇𝑜𝑡 𝐹𝑜𝑟 𝐷𝑜𝑚 𝑇𝑖,𝑡 = 𝑇𝑖,𝑡 + 𝑇𝑖,𝑡

)

(5a)

(5b)

(5c)

AC

CE

in which 𝐸𝑥𝑝𝑖,𝑡 represents the total export of electrical appliances by manufacturers in country i in year t, 𝑌𝑖,𝑡 the domestic production of electrical appliances in country i in year t, 𝐼𝑚𝑝𝑖,𝑗, the import in country i of electrical appliances manufactured in country j, Πki,t the patent stock in country i in year t of energy efficient innovations for appliance 𝑘 (k=1 for refrigerators and freezers, k=2 for washing machines, k=3 for dishwashers). Data on domestic production and bilateral trade flows (between our sample of countries and all other countries in the world), considered at an eight digit level of detail and expressed in monetary values (Euro), derive, respectively, from the PRODCOM and the COMEXT databases, both available from Eurostat (2016a, b). When calculating bilateral trade flows, for any importing country 𝑗, appliance 𝑘 and 𝑇𝑜𝑡 year 𝑡, we consider all the possible exporting countries worldwide. The variable 𝑇𝑖,𝑡 should be interpreted as the average number of energy efficiency patents (properly depreciated) that are embodied in a typical electrical appliance purchased by a consumer in country i and year t. [Error! Reference source not found. and Error! Reference source not found. about here] Error! Reference source not found. shows the share of the 'domestic' patent stock 𝐷𝑜𝑚 𝐷𝑜𝑚 𝐹𝑜𝑟 over total patents embodied in new electrical appliances (𝑇𝑖,𝑡 /(𝑇𝑖,𝑡 + 𝑇𝑖,𝑡 ) by country in year 1995 and 2013 whereas Error! Reference source not found. shows the 14

ACCEPTED MANUSCRIPT

CE

PT

ED

MA NU

SC

RI P

T

size of the patent stock (total, domestic and foreign) by country for the same years. Italy and Germany are the countries that relied the most on domestically-developed energy efficient innovations. More specifically, more than half of the energy efficient patents embodied in newly purchased appliances for Italian consumers in 1995 and German consumers in 2013 were developed, respectively, in Italy and Germany. Both countries are among the global leaders in the manufacturing of these appliances and also host important applicants in energy efficiency technologies (Costantini et al., 2015). Overall, what is interesting here is that accounting for technological change embodied in imported goods is crucial to evaluating the impact of technology on energy efficiency, since the foreign component generally represents a substantial share of adopted technologies. In fact, we may expect a relatively large improvement in technical efficiency also in technology-adopting countries since EE performances are not affected by the level of national or international technological capacity, but by the market penetration of new energy efficient appliances, no matter where these are manufactured. Another important issue is represented by potential policy spillover effects. Government regulation, in particular of policies that aim to promote energy saving, can strongly affect the incentive to consumers of a country to adopt EE technologies and the development of new technologies by both domestic and also foreign manufacturers of appliances. Costantini et al. (2017) point out that foreign countries characterized by greater innovation capacity have larger incentives to export new EE goods in those countries with higher policy stringency and a well-balanced mix of policy instruments, which results in enlarged market demand for more efficient goods. By construction, our innovation proxy also captures this indirect effect on foreign manufacturers since both domestic and foreign invention, measured by patents, is weighted by internal production and worldwide import flows. The resulting indicator has to be interpreted as the combined effect of both policy-induced and market-driven mechanisms aimed at triggering both the invention and adoption of energy efficient appliances. 3.3 Estimation

AC

The empirical contributions of SFA with panel data are mainly distinguishable in fixed and random effects models12. Fixed effects (FE) specifications allow unobserved heterogeneity to be captured among units of analysis, but the simple formulation of SFA fixed effects models treats the unit-specific inefficiency levels as fixed (Battese and Coelli, 1992; 1995; Kumbhakar, 1990; Pitt and Lee, 1981; Schmidt and Sickles, 1984, among others). This implies that the inefficiency term captures all the heterogeneity with no possibility of distinguishing between persistent actual inefficiency and timeinvariant heterogeneity. Different solutions have been proposed in order to separate technical inefficiency from other time invariant unobservable components. The True Fixed Effect (TFE) model introduced by Greene (2005a, b) addresses this issue by means of the maximum likelihood dummy variable method (MLDV). Even though it is computationally appealing, the TFE model suffers from two main limitations. First, the MLDV method implies a large number of parameters to be estimated (the unit-specific intercepts together with the structural parameters). In short panels, the incidental parameter problem (Neyman and Scott, 1948; Lancaster, 2002) leads to inconsistency in the variance estimation although the frontier coefficients remain unbiased. Belotti and Ilardi (2012) demonstrated that the inconsistency bias is negligible in panels with 12

For a review of SFA models, see Murillo-Zamorano (2004).

15

ACCEPTED MANUSCRIPT

AC

CE

PT

ED

MA NU

SC

RI P

T

𝑇 > 15. In this respect, two consistent estimators suitable for our analysis are represented by the ‘Marginal Maximum Likelihood Within’ estimator (MMLWE) by Chen et al. (2014) and the ‘Marginal Maximum Simulated Likelihood Estimator’ estimator (MMSLE) by Belotti and Ilardi (2015), both allowing a consistent estimation of the TFE model. A second major pitfalls of the TFE model is that the unit-specific intercepts capture all the time-invariant component so that the inefficiency can no longer be affected by any of the time invariant factors. This represents an extreme solution which does not actually solve the issue of unobserved heterogeneity in panel frontier analysis and may give rise to either upward and downward bias in efficiency scores depending on the nature of unobserved heterogeneity. SFA models also differ depending on the capacity to capture the transient or persistent part of technical efficiency. For instance, Colombi et al. (2014) and Tsionas and Kumbakhar (2014) have proposed the Generalized True Random Effects (GTRE) model to disentangle persistent technical inefficiency from short-run (or transient) inefficiency. A straightforward econometric estimation strategy of this model is presented by Filippini and Greene (2016). In this paper we use the TFE model by employing different estimators and, as a robustness check, the TRE model, to explain the variance of the inefficiency component through variables that do not enter directly the ‘production’ process. Specifically, the employed models allow the distribution moments of the inefficiency term to be modelled as a function of auxiliary variables. The stochastic frontier model requires both components of the error term (inefficiency and idiosyncratic error) to be independent of the vector of inputs. In our case, the inclusion of a technology proxy as an auxiliary variable is likely to be endogenous given its possible correlation with GDP per capita. Both theoretical and empirical research show that the level of affluence is affected by the rate and direction of technical change via productivity changes (Kumar and Russell, 2002, among others) as well as by the volume of trade (Grossman and Helpman, 1991; Perla et al., 2015). Moreover, several studies point to a significant role of policy regulations as key determinants of eco-innovations of which EE technologies constitute a large subset (Diaz-Rainey and Ashton, 2015; Costantini et al., 2017, among others). For this reason, the orthogonality assumption between the vector 𝑥 and the composite error term 𝜖𝑖,𝑡 would be violated, leading to biased estimation of the parameters of interest and the inefficiency component itself. In this respect, fixed effects models partially mitigate this potential endogeneity by controlling for time-invariant unobservables (Kumbhakar and Lovell, 2000). As a further mitigation device, we adopt the Two-Stage Residual Inclusion (2SRI) method developed by Terza et al. (2008)13. This approach includes a first stage obtained by regressing the endogenous variable, GPD per capita in our case, on one instrumental variable and all other exogenous covariates. As in the IV technique, the validity of the 2SRI method exclusively relies on the quality of instrument employed which has to be both relevant (i.e. correlated with the endogenous covariate) and exogenous with respect to both components of the error term. When identifying a potential search area for candidate instruments, we look at the 13

An alternative solution to allow endogeneity in non-linear models is represented by the Two-Stage Predictor Substitution (2SPS) in which fitted values instead of residuals are used in the second stage. However, Terza et al. (2008) demonstrated that the estimates deriving from the 2SPS are not consistent when non-linear models are employed.

16

ACCEPTED MANUSCRIPT

MA NU

SC

RI P

T

composition of public expenditure. In particular, we focus on the amount of per capita public health expenditure as an instrument to be employed in our first stage estimation. The economic and statistical relevance of this instrument emerges from a large number of empirical works that compare the growth of GDP and the associated level of health spending. In particular, by controlling for other potential drivers such as the population’s structure and technological progress, these studies find strong significant association between health expenditure and GDP (Baltagi and Moscone, 2010; HaliciTuluce et al., 2016; Hitiris and Posnett, 1992; Kleiman, 1974; Newhouse, 1977). Building on this consistent evidence, we argue that the amount of health expenditure represents a good predictor of GDP over space and time. We also argue that this instrument is able to capture exogenous variations of GDP per capita that go beyond productivity improvements due to the technological capacity of the country and can thus be fruitfully employed in our second-stage estimates14 in which first-stage residuals ̂𝑖,𝑡 ) enter the EDSFM as an additional covariate. (𝐶𝐹 Consistently with our empirical strategy, the technology-augmented EDSF-TFE model is as follows: ̂𝑖,𝑡 + 𝛽2 ln(𝑃_𝑒𝑙𝑦𝑖,𝑡 ) + 𝛽3 ln(𝐺𝐷𝑃𝑖,𝑡 /𝑃𝑜𝑝𝑖,𝑡 ) + ln(𝐸𝑖,𝑡 ) = 𝛼𝑖 + 𝛽1 𝐶𝐹

ED

+ 𝛽4 ln(𝐷𝑤𝑒_𝑠𝑖𝑧𝑒𝑖,𝑡 ) + 𝛽5 𝑆ℎ_𝑢𝑟𝑏𝑎𝑛_𝑝𝑜𝑝𝑖,𝑡 + 𝑣𝑖,𝑡 + 𝑢𝑖,𝑡

(6)

(7)

ln(𝑃𝑜𝑙_𝑖𝑛𝑑𝑒𝑥𝑖,𝑡 ) 𝓏𝑖,𝑡 = [ ] 𝑇𝑜𝑡 ln(𝑇𝑖,𝑡 )

(8)

CE

PT

′ 𝑢𝑖,𝑡 = 𝑧𝑖,𝑡 𝜓 + 𝜂𝑖,𝑡

4 Results and discussion

AC

Table 2 reports the results obtained from the EDSF-TFE model with residual inclusion15. We present four specifications. Specification (1) shows the results of the EDSF-TFE model without modelling the inefficiency term. Specification (2) accounts for the possible role of demand-pull policies as a driver of technical efficiency improvement whereas in specification (3) we also include our indicator of technology market penetration of energy efficient appliances in addition to the policy variable. Finally, in specification (4), we also evaluate whether relying on foreign knowledge (embodied in imported appliances) has an impact on energy efficiency. Model (1) is estimated with the ‘within’ estimator and models (2)-(4) are estimated by means of the

14

First-stage estimates show that our instrument is strong enough and goes in the expected direction. The coefficient of per-capita health expenditure is 0.559 (t-stat 7.11) and its correlation with GDP is 72%. The F-statistics of excluded instruments is 50.53. 15 Standard Student's t statistics (in parenthesis in Table 2) do not account for the fact that residuals from the first stage are estimated. In Table 2 we also report Student’s t statistics estimated with bootstrap replications (200 replications) in brackets. In addition to estimates that account for the endogeneity of GDP per capita, we also report (as a benchmark) results that do not account for the endogeneity of GDP per capita in Table 3 (column 1).

17

ACCEPTED MANUSCRIPT MMSL.16 The inefficiency component is assumed to be half-normally distributed. [Table 2 about here]

AC

CE

PT

ED

MA NU

SC

RI P

T

It all the estimated models, the size of the parameter lambda17, calculated as the ratio of the standard deviation of the inefficiency component to the standard deviation of the idiosyncratic component, ranges from 1.7 in column 1) to 0.38 in column 3); it is statistical significant at 1 percent level in columns 2), 3) and 4) and significant at the 10 percent level in column 1). The price elasticity of the demand for electricity is negative and significantly different from zero. The range of variation of point coefficients (between -0.23 and -0.27, depending on the model specification) is fully consistent with results found in the existing literature that focuses on the residential sector (Filippini et al., 2014; Alberini and Filippini, 2011, among others). The effect of GDP per capita on the frontier of electricity consumption is negative and statistically significant in all specifications. Increases in average GDP per capita within the country are correlated with a ceteris paribus reduction in the demand for energy services provided by electrical appliances. Even though this result appears somehow unexpected, it is in line with empirical findings which suggest that the residential consumption of electricity is an inferior good in wealthy countries in the long run (Fullerton et al., 2012; Narayan et al., 2006). This means that beyond a certain point, marginal increases in wealth result in a reduction in the demand for basic energy services. Among the different energy services in the residential sector, in the case of traditional electrical appliances for washing and cooling, the contribution of allocative efficiency (i.e. how household combine different inputs to obtain energy services) to energy saving is limited and most of the electricity reduction is due to gains in technical efficiency which in turn depend on the level of technology embodied in new appliances captured by our innovation proxy. As far as other control variables are concerned, average dwelling size turns out to be statistically insignificant in all specifications with exception of the one of column 4, where we estimate a negative effect (p<0.1, but insignificant with bootstrap standard errors): larger dwellings may benefit from some (weak) economies of scale in the provision of energy services. On the other hand, the indicator of urbanization is negative and generally statistically significant, suggesting that increased urbanization leads to a decrease in the demand for traditional energy services (cooling and washing) due to behavioural changes. Residuals from the first stage significantly influence the stochastic frontier estimation. Since the statistical significance of residuals from the first stage represents an explicit test for endogeneity (where the null hypothesis is that the suspect endogenous covariate is exogenous, see Terza et al, 2008), these results suggest that GDP per capita is endogenous even when accounting for time-invariant heterogeneity across countries18. Estimates are performed using the software StataTM and the user-written command sftfe by Belotti and Ilardi (2015). 17 When sfpanel and sftfe are used to estimated heteroschedastic models, the standard errors of the lambda parameter are not automatically calculated. We compute t-statistics with bootstrap replications. 18 As a robustness check we repeat our analysis without accounting for the potential endogeneity of GDP per capita. Results are discussed later in Section 4.1. 16

18

ACCEPTED MANUSCRIPT

CE

PT

ED

MA NU

SC

RI P

T

When we look at the drivers of variance of the inefficiency term, we observe that in column 2 the policy index is negatively correlated with technical inefficiency, the effect being statistical significant only when inference is done with bootstrap. This finding, that is in line with the existing literature, suggests that policies contribute to the improvement in energy efficiency of domestic appliances. However, when we add our indicator of technology to the policy index (column 3), the impact of EE policies disappears (i.e. the point coefficient becomes positive but not statistically significant) whereas the impact of EE technologies turns out to be negative and significant. This evidence shows that inclusion of the innovation proxy provides a more complete representation of market dynamics in the sector of large traditional electrical appliances since this indicator is able to capture both the market-demand policy stimulus and other direct and indirect mechanisms aimed at boosting the replacement of old appliances with newer ones. Our indicator shapes technological advancement as a result of firms’ exploitation of new technology markets deriving from both endogenous profitmaximization behaviour and technology-push incentives, trade liberalization and adoption dynamics. When we include the share of domestic patents to disentangle the contribution of domestic and foreign technology components (column 4), the former shows a positive sign in explaining the variance of technical inefficiency. This result suggests that gaining access to advanced technologies developed abroad is crucial to improve domestic energy efficiency in home appliances. When calculating the marginal effects of the two main efficiency drivers we estimate that a 1 percent increase in the intensity of the policy indicator results in a reduction in the technical inefficiency of 0.11 percent (column 2, Table 2) while a 1 percent increase in the technology stock results in a 0.43 percent decrease in the technical inefficiency. These figures confirm the larger effects due to the role of technology invention and adoption when compared to our policy indicator. 4.1 Robustness checks

AC

A first robustness check of our main results (using column 4 of Table 2 as our favourite specification) involves an estimate that does not account for the endogeneity of GDP per capita, to be used as a benchmark. A potential explanation for the exogeneity of GDP with regard to our innovation proxy lies with the specific nature of the technologies we consider. EE technologies embodied in traditional electrical appliances may be more functional to economic growth in some export-intensive countries and the modelling of adoption dynamics by means of worldwide bilateral trade flows is likely to increase the correlation between GDP and innovation dynamics. However, the potential bias may be not very large since in the set of EU countries analysed the total value of trade considered here (i.e. domestic appliances) represents only a very small fraction of total GDP 19. Since IV estimates in the presence of exogenous variables deliver inefficient results (i.e. with larger standard errors than in the non-IV case), in column 1 of Table 3 we show the results for the simple case in which GDP per capita is considered exogenous. The results for the estimation of the stochastic frontier remain basically unaffected as far as their sign are considered. Some difference from our baseline results is observed, however, in the magnitude and 19

The amount of import value of large electrical appliances for EU-28 in 2013 corresponds to 0.018 percent of total GDP.

19

ACCEPTED MANUSCRIPT

PT

ED

MA NU

SC

RI P

T

statistical significance of the estimated effects. The magnitude of the estimated impact of technology on the variance of inefficiency is about half as large as the one in column (4) of Table 2 and it is only weakly significant (p<0.1). Moreover, the positive impact of relying on domestic knowledge on the variance of the inefficiency component turns out to be statistically insignificant. These differences suggest to account for the endogeneity of GDP per capita. As a second robustness check, we re-estimate our favourite specification (column 4 of Table 2) by assuming that the technical inefficiency component follows the exponential rather than the half normal distribution. Results are shown in column 2 of Table 3. Overall, the results are confirmed in sign, magnitude and statistical significance. The only difference is that while the standard errors that do not account for the first step of the IV approach (in parenthesis) suggest that technology and ‘domestic share’ effects are statistical significant, standard errors estimated with bootstrap result in insignificant coefficients (p-value=0.14 for technology, p-value=0.24 for the share of domestic patents). Finally, as discussed in Section 3.3, a further robustness check is provided by showing estimates with the True Random Effect model (Table 3).20 Overall, our main results are confirmed. Some change, however, can be seen in the magnitude of the estimated effects. As far as the variables that explain the variance of the inefficiency component are concerned, there is a substantial increase in the magnitude of the coefficient of our technology-related variables. It all specifications reported in Table 3, the parameter lambda is always significant at 1% level, with a magnitude ranging from 0.21 to 0.93. [Table 3 about here]

CE

4.2 Efficiency scores

AC

In Figure 5 we report estimated technical efficiency scores in 1995 and 2013 for various specifications of stochastic frontier: TFE results with no auxiliary variable in (results from column 1 of Table 2, reported in the top-left panel), TFE results with policy, technology and share of domestic patents included as auxiliary variables (results from column 4 of Table 2, reported in the top-right panel) and TRE with no auxiliary variables (results from column 3 of Table 3, reported in the bottom-left panel).

[Figure 5 about here] Overall, estimated technical efficiency scores are very large for all different specifications and are always above 60 percent, with an average value of 95.7 percent (across different years, estimators and specifications). These high efficiency scores are consistent with the fact that we are focusing our attention on transient technical inefficiency and are in line with similar results for the residential sector reported in The true random effect model was estimated with the software StataTM by means of the user-written command sfpanel, developed by Belotti et al. (2015). 20

20

ACCEPTED MANUSCRIPT

MA NU

SC

RI P

T

Filippini et al. (2014). We observe that Austria appears to be the best performing country in terms of the efficiency score in 1995 (efficiency greater than 98 percent for all specifications). Evidences for top performing countries in 2013 appear more mixed: in the top and bottom left panels of Figure 5 we observe the UK, Italy and Denmark to be the three best performing countries. In our favourite specification (top-right panel of Figure 5), however, we estimate a technical efficiency above 99 percent for all countries but Germany (98.7 percent), that makes rankings uninformative. The worst performing country in 1995 was Sweden in our favourite specification (efficiency score of 86 percent), Germany for the specification where we do not model the variance of the inefficiency term (92 percent) and the UK for the TRE model (76 percent). The worst performing country in 2013 was Greece with an estimated efficiency score that ranges between 69 percent (bottom-left panel of Figure 5) and 88 percent (top-left panel of Figure 5). UK was the country that, on average, worsened its technical efficiency the most in the considered period whereas the average largest increase in technical efficiency is found for Greece. [Table 4 about here]

CE

5 Conclusions

PT

ED

Given the different nature of the models employed, we also provide a consistency check for all the efficiency specifications. Table 4 shows the between-countries correlation matrix for 1995 and 2013. The correlations are always positive: correlation is high between efficiency scores estimated by the TRE and TFE specifications that include the full set of auxiliary variable explaining the variance of inefficiency (almost 99 percent) and smaller between these estimates and the ones derived from the specification that does not include any auxiliary variable (about 34-35 percent).

AC

This study presents an original methodology for accounting for the role of innovation dynamics in the framework of SFA, a well-known parametric technique that is able to disentangle technical efficiency, in our case transient inefficiency, as a measure of distance between the observed and the maximum theoretically efficient frontier of electricity consumption. The approach proposed here is tested by analysing the efficiency trend in two groups of traditional home electrical appliances, also called ‘white goods’, in the period 19952013 in ten European countries. The choice of using domestic appliances aimed at fulfilling primary needs such as cooling or washing which show a low behavioural response of consumers to changes in energy prices, minimizes the potential impact of the rebound effect and helps to better identify the specific impact of technology market penetration in reducing energy consumption. Building on the growing empirical literature on eco-innovation, patent applications in specific technology classes are employed to identify EE technologies embodied in the set of considered appliances, namely freezers and refrigerators, washing machines and dishwashers. As a first step, we take into account the growing regulatory activity existing in Europe that aims to boost EE in the residential appliances sector. Our policy indicator includes the cumulated count of the regulations simultaneously implemented in each country and year. However, this approach mostly explains only demand-pull drivers and does not account for all the different market forces that shape the renovation

21

ACCEPTED MANUSCRIPT

AC

CE

PT

ED

MA NU

SC

RI P

T

process of older appliances. Given the variety of actors and processes involved in the market of white goods, both on the supply and demand side, a more comprehensive approach is needed to model the combined effects of all these forces. Our innovation proxy fills this gap by capturing both firms’ design and development of new technologies (invention) and market drivers able to determine the adoption level of the different models which include EE policies, trade barriers and other drivers that determine the technology penetration rate of new efficient equipment. Patent information on EE technologies for the group of goods considered are combined with specific data on values for both domestically manufactured and imported (worldwide) goods which allows us to build a reliable proxy of the level of market penetration of new energy efficient appliances in the national residential market. By explicitly considering the net volumes of new equipment on the market deriving both from domestic production and import, our proxy of technology penetration captures the impacts of a series of direct and indirect elements in the value chain of traditional electrical appliances. First, it considers the entire pool of national implemented policy instruments that aim to boost the replacement rate of older appliances. By accounting for international trade, we also consider countries in which the internal production of appliances is negligible. These countries are likely to adopt a policy setting that is particularly favourable towards increasing EE, with strong demand-pull interventions that intend to sustain internal demand. In the absence of trade barriers, foreign manufactures are likely to redirect their export capacity to match the policy-induced households' demand for efficient appliances in these countries. We take advantage from the existing literature on the derived households’ energy demand in order to fruitfully employ a policy- and technology-augmented specification of stochastic frontier and associated efficiency scores. Moreover, the 2SRI approach proposed by Terza et al (2008) and here introduced in the SFA framework allows to control for potential endogeneity which may arise when innovation processes and economic growth are considered in the frontier specification. In order to test the effectiveness of our method, we compare the policy-augmented EDSFM specification as proposed by Filippini et al. (2014) with the same model in which our innovation proxy is added to the policy indicator as a driver of technical inefficiency. Results for policies turn out to be no longer significant and are absorbed by our innovation proxy. As a robustness check, we run the same specification with different estimators, all confirming our results, which imply substantive policy implications. First, the combination of different actors engaged in the value chain of the traditional home appliances market generates mechanisms that cannot be fully explained by demand-pull mechanisms. For instance, policies implemented in the EU in the large traditional appliances sector are important drivers of consumption reduction, but their effects appear to be less significant in explaining the technical efficiency level. On the contrary, other market dimensions, such as technology competition and market penetration, are major determinants of efficiency gains. In this regard, a further policy stimulus towards supply-push drivers such as innovation policies for EE-oriented R&D activities and interventions that aim to reduce trade barriers and increase the manufacturers’ network capacity (through ICTs investments or collaboration projects) would be key complementary actions to policies targeted at consumers. With regard to the efficiency scores, our estimates indicate a range from around 70 percent to almost 100 percent, depending on the country, year, model specification and

22

ACCEPTED MANUSCRIPT

AC

CE

PT

ED

MA NU

SC

RI P

T

estimator employed. This evidence suggests that households are highly efficient in combining ‘energy inputs’ at minimum cost in order to obtain energy services such as cooling and washing. However, these figures also reflect the technology maturity of traditional appliances and the fact that households have a limited range of action in changing the way they use large traditional appliances which operate continuously most of the time. Although our results provide interesting insights into the link between invention and innovation of energy efficient appliances, their adoption by consumers and their contribution to improved energy efficiency, some limitations of the analysis should be highlighted, leaving room for further research. First, we focus on the transient part of technical efficiency, while the persistent one remains unaddressed with the class of empirical estimators here employed. Moreover, our limited sample of countries reduces the external validity of the results, also considering that the pool of analysed countries is biased towards Western EU countries where markets are more efficient, households are wealthier and policy makers have set the EE gain as a priority for several years.

23

ACCEPTED MANUSCRIPT

Acknowledgements

AC

CE

PT

ED

MA NU

SC

RI P

T

We are indebted to two anonymous referees for their insightful comments which have helped to substantively enrich the initial version of the paper. We also thank Giuseppe Ilardi (Bank of Italy) and participants in the first IAERE conference held in 2014 at the Department of Economics at Ferrara University. Financial support from the Italian Ministry of Education, University and Research (Scientific Research Program of National Relevance 2010 on “Climate change in the Mediterranean area: scenarios, economic impacts, mitigation policies and technological innovation”) is gratefully acknowledged. The usual disclaimer applies.

24

ACCEPTED MANUSCRIPT

References

AC

CE

PT

ED

MA NU

SC

RI P

T

Aigner D., Lovell C., Schmidt P. (1977). Formulation and estimation of stochastic frontier production function models. Journal of Econometrics, Vol. 6, pp. 21-37. Alberini A., Filippini M. (2011). Response of residential electricity demand to price: The effect of measurement error. Energy Economics, Vol. 33, pp. 889-895. Ang B.W. (1995). Multilevel decomposition of industrial energy consumption. Energy Economics, Vol. 17(1), pp. 39-51. Baltagi Badi H., Moscone F. (2010). Health Care Expenditure and Income in the OECD Reconsidered: Evidence from Panel Data. Economic Modelling, Vol. 27, pp. 804811. Barbieri N., Palma A., (2017). Mapping energy-efficient technological advances in home appliances. Energy Efficiency, Vol. 10, pp. 693-716. Battese G., Coelli T. (1992). Frontier production functions, technical efficiency and panel data: with application to paddy farmers in India. Journal of Productivity Analysis, Vol. 3(1-2), pp. 153–169. Battese G., Coelli T. (1995). A model for technical inefficiency effects in a stochastic frontier production function for panel data. Empirical Economics, Vol. 20, pp. 325– 332. Belotti F., Ilardi G. (2012). Consistent estimation of the “true” fixed-effects stochastic frontier model. CEIS Research Papers (231). Belotti F., Ilardi G. (2015). Consistent Inference in Fixed-effects Stochastic Frontier Models. Temi di discussione (Economic working papers), Bank of Italy (forthcoming). Belotti F., Daidone S., Ilardi G., Atella V. (2013). Stochastic frontier analysis using Stata. The Stata Journal, Vol. 13(4), pp. 719-758. Bertoldi P., López-Lorente J., Labanca N. (2016). Energy Consumption and Energy Efficiency Trends in the EU-28 2000-2014. Scientific Reports of the Joint Research Centre, European Commission, JRC101177. Bigano A., Ortiz R. A., Markandya A., Menichetti E., Pierfederici R. (2010). The linkages between energy efficiency and security of energy supply in Europe. FEEM Nota di Lavoro 64/2010. Boyd, G. A. (2008). Estimating Plant Level Energy Efficiency with a Stochastic Frontier. The Energy Journal, Vol. 29, pp. 23-43. Brown M. L. (2004) Obstacles to Energy Efficiency, Vol. 4. pp. 465-475. In Encyclopedia of Energy, ed. by Cutler J. Cleveland, Elsevier, New York. Buck J., Young D. (2007). The potential for energy efficiency gains in the Canadian commercial building sector: A stochastic frontier study. Energy, Vol. 32, pp. 17691780. Burwell C.C., Sweezey B.G. (1990). The home: evolving technologies for satisfying human wants. In: Schurr S.H.; Burwell C.C.; Devine W.D.; Sonenblum S. (eds), Electricity in the American economy: agent of technological progress, Greenwood Press, New York. Cabeza L. F., Urge-Vorsatz D., McNeil M. A., Barreneche C., Serrano S. (2014). Investigating greenhouse challenge from growing trends of electricity consumption through home appliances in buildings. Renewable and Sustainable Energy Reviews, 25

ACCEPTED MANUSCRIPT

AC

CE

PT

ED

MA NU

SC

RI P

T

Vol. 36, pp. 188-193. Chakravarty D. Dasgupta S., Roy J. (2013). Rebound effect: how much to worry? Current Opinion in Environmental Sustainability, Vol. 5, pp. 216-228. Chen Y., Wang H., Schmidt P. (2014). Consistent estimation of the fixed effects stochastic frontier model. Journal of Econometrics, Vol. 181, pp. 65-76. Chitnis M., Sorrell S. (2015). Living up to expectations: estimating direct and indirect rebound effects for UK households. Energy Economics, Vol. 52 pp. S100-S116. Colombi R., Kumbhakar S. C., Martini G., Vittadini G. (2014). Closed-skew normality in stochastic frontiers with individual effects and long/short-run efficiency. Journal of Productivity Analysis, Vol. 42(2), pp. 123-136. Costantini V., Liberati P. (2014). Technology transfer, institutions and development. Technological Forecasting and Social Change, Vol. 88, pp. 26-48. Costantini V., Crespi F., Palma A. (2017). Characterizing the policy mix and its impact on eco-innovation in energy-efficient technologies. Research Policy, Vol. 46(4), pp. 799-819. Costantini V., Crespi F., Palma A. (2015). Mapping innovation systems through patent analysis: the case of energy efficiency in the residential sector. In The Economics of Knowledge Generation and Distribution: The Role of Interactions in the System Dynamics of Innovation and Growth (Ed. Paolo Patrucco). Routledge, London. Datta S., Gulati S. (2014). Utility rebates for ENERGY STAR appliances: Are they effective? Journal of Environmental Economics and Management, Vol. 68 (3), pp. 480-506. Davis L. (2008). Durable Goods and Residential Demand for Energy and Water: Evidence from a Field Trial. RAND Journal of Economics, Vol. 39 (2), pp. 530-546. de Miguel C., Labandeira X., Loschel A (2015). Frontiers in the Economics of Energy Efficiency. Energy Economics, Vol. 52 pp. S1-S4. del Río P., Carrillo-Hermosilla J., Könnölä, T. (2010), Policy Strategies to Promote Eco-Innovation. Journal of Industrial Ecology, Vol. 14, pp. 541-557. Diaz-Rainey I., Ashton J. K. (2015). Investment inefficiency and the adoption of ecoinnovations:The case of household energy efficiency technologies. Energy Policy, Vol. 82 pp. 105-117. Ek K., Söderholm P. (2010). The devil is in the details: Household electricity saving behavior and the role of information. Energy Policy, Vol. 38(3), pp. 1578-1587. ETC-SCP (European Topic Centre on Sustainable Consumption and Production) (2013). Housing assessment, final report. Task 2.5.1.1. ETC/SCP Working Paper No 4. European Commission (2011). COM (2011) 109 final. Communication from the Commission to the European Parliament, the Council, the European Economic and Social Committee and the Committee of the Regions. Energy Efficiency Plan 2011. Brussels. European Commission (EC) (2012a). Energy Efficiency Status Report 2012. Electricity Consumption and efficiency Trends in the EU-27. Joint Research Centre, Institute for Energy and Transport. European Commission (EC) (2012b). “Energy Efficiency Directive”, 2012/27. European Commission (EC) (2014). A policy framework for climate and energy in the

26

ACCEPTED MANUSCRIPT

AC

CE

PT

ED

MA NU

SC

RI P

T

period from 2020 to 2030. Com.(2014) 15 Final, Brussels. Eurostat (2016a). Sold production, exports and imports by PRODCOM list (NACE Rev. 2) - annual data. Available at: http://ec.europa.eu/eurostat/web/prodcom/data/database. Eurostat (2016b). EU trade statistics since 1988 by CN8. Available at: http://ec.europa.eu/eurostat/web/international-trade/data/database. Evenson R.E. (2002). Induced adaptive invention/innovation and productivity convergence in developing countries. In Technological Change and the Environment. Eds. Grubler A., Nakicenovic N., Nordhaus W.D. (2002). Washington, DC. Resource Future Ref. 131, pp. 61-96. Fernandez V.P. (2001). Observable and unobservable determinants of replacement of home appliances. Energy Economics, 23. pp. 305-323. Filippini M. (1995). Swiss residential demand for electricity by time-of-use. Resource and Energy Economics, Vol. 17(3), pp. 281-290. Filippini M., Hunt L. C. (2012). US residential energy demand and energy efficiency: A stochastic demand frontier approach. Energy Economics, Vol. 34(5), pp. 1484-1491. Filippini M., Hunt L. C. (2015). Measurement of energy efficiency based on economic foundations. Energy Economics, Vol. 52(supp. 1), pp. 5-16. Filippini M., Pachauri S. (2004). Elasticities of electricity demand in urban Indian households. Energy Policy, Vol. 32(3), pp. 429-436. Filippini, M. Hunt L. C., Zorić J. (2014). Impact of energy policy instruments on the estimated level of underlying energy efficiency in the EU residential sector. Energy Policy, Vol. 69(C), pp. 73-81. Filippini M., Geissmann T., Greene W. H. (2016). Persistent and Transient Cost Efficiency – An Application to the Swiss Hydropower Sector. CER-ETH – Center of Economic Research at ETH Zurich Working Paper No. 16/251. Filippini M., Greene W. (2016). Persistent and transient productive inefficiency: a maximum simulated likelihood approach. Journal of Productivity Analysis, Vol. 45 pp. 187-196. Fullerton Jr T. M., Juarez D. A., Walke A. G. (2012). Residential electricity consumption in Seattle. Energy Economics, Vol. 34(5), pp. 1693-1699. Gans W., Alberini A., Longo A. (2013). Smart meter devices and the effect of feedback on residential electricity consumption: Evidence from a natural experiment in Northern Ireland. Energy Economics, Vol. 36. pp. 729-743. Geroski P. A. (2000). Models of technology diffusion. Research Policy, Vol. 29(4-5), pp. 603-625. Gillingham K., Palmer K. (2014). Bridging the Energy Efficiency Gap: Policy Insights from Economic Theory and Empirical Evidence. Review of Environmental Economics and Policy, 8(1), pp. 18-38. Greene W. (2005a). Reconsidering heterogeneity in panel data estimators of the stochastic frontier model. Journal of Econometrics, Vol. 126, pp. 269–303. Greene W. (2005b). Fixed and random effects in stochastic frontier models. Journal of Productivity Analysis, Vol. 23, pp. 7–32. Greening L. A., Greene D. L., Difiglio C. (2000). Energy efficiency and consumption the rebound effect - a survey. Energy Policy, Vol. 28 pp. 389-401. 27

ACCEPTED MANUSCRIPT

AC

CE

PT

ED

MA NU

SC

RI P

T

Griliches Z. (1957). Hybrid Corn: An Exploration in the Economics of Technological Change. Econometrica, Vol. 25(4), pp. 501-522. Griliches Z. (1998). Patent Statistics as Economic Indicators: A Survey. In R&D and Productivity: The Econometric Evidence. Ed. Griliches Z. National Bureau of Economic Research. University of Chicago Press. Grossman Gene M., Helpman Elhanan (1991). Trade, knowledge spillovers, and growth, European Economic Review, vo. 35(2), pp. 517-526. Halıcı-Tülüce, N.S., Doğan, İ., Dumrul, C. (2016). International Journal of Health Economics and Management, Vol. 16(1), pp.23-49. Hall B. H. (2007). Measuring the returns to R&D: the depreciation problem. NBER Working Paper 13473. Hall B. H., Jaffe A., Trajtenberg M. (2005). Market Value And Patent Citations. Rand Journal of Economics Vol. 36 (1), pp. 16-38. Hartman, R S. (1979). Frontiers in Energy Demand Modeling. Annual Review of Energy, Vol. 4, pp. 433-466. Hitiris T., Posnett J. (1992). The determinants and effects of health expenditure in developed countries, Journal of Health Economics, Vol. 11(2), pp. 173-181. IEA (International Energy Agency) (2009). Gadgets and Gigawatts. Policies for Energy Efficient Electronics. OECD-IEA Publishing, Paris. IEA (International Energy Agency) (2013). Transition to sustainable buildings. Strategies and opportunities to 2050. International Energy Agency, Paris. Jaffe A, B., Newell R. G., Stavins R. N.(2004). Economics of Energy Efficiency, In Encyclopedia of Energy, edited by Cutler J. Cleveland, Elsevier, New York. Jaffe A. B., Stavins R. N. (1994). The energy-efficiency gap What does it mean? Energy Policy, 22(10), 804-810. Jaffe A. B., Stavins R. N. (1995). Dynamic Incentives of Environmental Regulations: The Effects of Alternative Policy Instruments on Technology Diffusion. Journal of Environmental Economics and Management. Vol. 29 (3), pp. S43–S63. Jaffe A., Trajtenberg M. (2004). Patents, Citations, and Innovations: A Window on the Knowledge Economy. Journal of Economic Literature, 42 (4), pp. 1158-1160. Johnstone N., Haščič I., Popp D. (2010). Renewable Energy Policies and Technological Innovation: Evidence Based on Patent Counts. Environmental and Resource Economics, Vol. 45(1), pp 133-155. Karshenas M., Stoneman P. L. (1993). Rank, stock, order, and epidemic effects in the diffusion of new process technologies: an empirical model. RAND Journal of Economics 24, 503-528. Kavousian A., Rajagopal R., Fischer M. (2013). Determinants of residential electricity consumption: Using smart meter data to examine the effect of climate, building characteristics, appliance stock, and occupants’ behaviour. Energy, Vol. 55, pp. 184194. Kaza N. (2010). Understanding the spectrum of residential energy consumption: a quantile regression approach. Energy Policy, Vol. 38, pp. 6574-6585. Khazzoom J. D. (1980). Economic implications of mandated efficiency in standards for household appliances. Energy Journal, Vol. 1(4), pp. 21-40. Kleiman, E. (1974). The Determinants of National Outlay on Health‖, in: Perlman, M. 28

ACCEPTED MANUSCRIPT

AC

CE

PT

ED

MA NU

SC

RI P

T

(Ed.), The Economics of Health and Medical Care, Macmillan, London. Kumar, S.; R. Russell (2002). Technological Change, Technological Catch-up, and Capital Deepening: Relative Contributions to Growth and Convergence. American Economic Review, Vol. 92, pp. 527-548. Kumbhakar S. C. (1990). Production frontiers, panel data and time-varying technical inefficiency. Journal of Econometrics, Vol. 46, pp. 201–212. Kumbhakar S. C., Lovell C. A. K. (2000). Stochastic frontier analysis. Cambridge University Press. Lancaster T. (2002). The incidental parameters problem since 1948. Journal of Econometrics, Vol. 95, pp. 391-414. Lanjouw J., Pakes A., Putnam J. (1998). How to Count Patents and Value Intellectual Property: Uses of Patent Renewal and Applications Data. Journal of Industrial Economics, 46 (4), pp. 405-433. Linares P., Labandeira X. (2010). Energy Efficiency: Economics And Policy. Journal of Economic Surveys, Vol. 24(3), pp.573-592. Malerba F., Orsenigo L. (1996). The dynamics and evolution of industries. Industrial and Corporate Change, 5 (1), pp. 51-87. McKinsey (2009) Pathways to a low-carbon Economy: Version 2 of the Global Greenhouse Gas Abatement Cost Curve; McKinsey & C. McKinsey (2012). Battle for the home of the future: How utilities can win. In McKinsey on Sustainability and Resource Productivity No. 1, Summer 2012 McKinsey & Company Industry Publications. Meeusen W., van den Broeck J. (1977). Efficiency estimation from Cobb-Douglas production function with composed errors. International Economic Review, Vol. 18, pp. 435-444. Mills B., Schleich J. (2010). What's driving energy efficient appliance label awareness and purchase propensity? Energy Policy, Vol. 38(2), pp. 814-825 Murillo-Zamorano (2004). Economic efficiency and frontier techniques. Journal of Economic Surveys, Vol. 18 (1), pp. 33-78. Narayanm P. K., Smyth R., Prasad A.(2007). Electricity consumption in G7 countries: A panel cointegration analysis of residential demand elasticities. Energy Policy, Vol. 35(9), pp. 4485-4494. Nesta L., Vona F., Nicolli F. (2014). Environmental policies, competition and innovation in renewable energy. Journal of Environmental Economics and Management, Vol. 67(3), pp. 396-411. Newell R. G., Jaffe A., Stavins R.N. (1999). The Induced Innovation Hypothesis and Energy-Saving Technological Change. The Quarterly Journal of Economics, Vol. 114 (3), pp. 941-975. Newhouse J.P. (1977). Medical Care Expenditure: A Cross-National Survey. Journal of Human Resources, Vol. 12, pp. 115-125. Neyman J., Scott E. (1948). Consistent estimates based on partially consistent observations. Econometrica, Vol. 16, pp. 1–32. Noailly J. (2012). Improving the energy efficiency of buildings: The impact of environmental policy on technological innovation. Energy Economics, Vol. 34 (3) pp. 795-806.

29

ACCEPTED MANUSCRIPT

AC

CE

PT

ED

MA NU

SC

RI P

T

OECD (2003). Environmentally Sustainable Buildings: Challenges and Policies. OECD, Paris. Orea L., Llorca M., Filippini M. (2015) A new approach to measuring the rebound effect associated to energy efficiency improvements: An application to the US residential demand. Energy Economics, Vol. 49 pp. 599-609. Perla J., Tonetti C., Waugh M. E. (2015). Equilibrium technology diffusion, trade, and growth. NBER Working paper no. 20881. Pitt M., Lee L. (1981). The measurement and sources of technical inefficiency in the Indonesian weaving industry. Journal of Development Economics Vol. 9, pp. 43–64 Popp D. (2002). Induced Innovation and Energy Prices. American Economic Review, 92(1), pp.160-180. Popp D. (2003). Pollution control innovations and the Clean Air Act of 1990. Journal of Policy Analysis and Management, Vol. 22(4), pp. 641-660. Popp D. (2006). International Innovation and Diffusion of Air Pollution Control Technologies: The Effects of NOX and SO2 Regulation in the US, Japan, and Germany. Journal of Environmental Economics and Management, Vol. 51(1), pp.4671. Ramos A., Gago A., Labandeira X., Linares P. (2015). The role of information for energy efficiency in the residential sector. Energy Economics, Vol. 52 pp. S17-S29. Saunders H. (2013). Historical Evidence for Energy Consumption Rebound in 30 US Sectors and a Toolkit for Rebound Analysts. Technological Forecasting and Social Change, Vol. 80 (7), pp. 1317-1330. Schleich J., Klobasa M., Gölz S., Brunner M., (2013). Effects of feedback on residential electricity demand. Findings from a field trial in Austria. Energy Policy, Vol. 61, pp. 1097-1106. Schmidt P., Sickles R. C. (1984). Production frontiers and panel data. Journal of Business Economics and Statistics, Vol. 2 pp. 367-374. Shih H.-Y., Chang T.-L.S. (2009). International diffusion of embodied and disembodied technology: a network analysis approach. Technological Forecasting and Social Change, Vol. 76, pp. 821-834. Sorrell S., Dimitropoulos J. (2008). The rebound effect: Microeconomic definitions, limitations and extensions. Ecological Economics, Vol. 65 (3), pp. 636-649. Sovacool B. K. (2009). The importance of comprehensiveness in renewable electricity and energy-efficiency policy. Energy Policy, Vol. 37(4), pp. 1529-1541. Stern D. I. (2012). Modeling international trends in energy efficiency. Energy Economics, Vol. 34(6), pp. 2200-2208. Stoneman P. (1983). The Economic Analysis of Technological Change. Oxford University Press, Oxford. Stoneman P. (2001). The Economics of Technological Diffusion. Wiley-Blackwell. Terza J. V., Basu A., Rathouz P. J. (2008). Two-Stage Residual Inclusion Estimation: Addressing Endogeneity in Health Econometric Modeling, Journal of Health Economics, 27(3), pp. 531-543. Thomas S. (2015). Energy efficiency policies for appliances. Report. Wuppertal Institute for Climate, Environment and Energy. Thore S., Kozmetsky G., Phillips F. (1994). DEA of financial statement data: the U.S. 30

ACCEPTED MANUSCRIPT

AC

CE

PT

ED

MA NU

SC

RI P

T

computer industry. Journal of Productivity Analysis, Vol. 5, pp. 229-248. Tsionas E. G.; Kumbhakar S. C. (2014). Firm heterogeneity, persistent and transient technical inefficiency: a generalized true random-effects model. Journal of Applied Economics, Vol. 29, pp. 110-132. van den Bergh J. C. J. M. (2011). Energy Conservation More Effective with Rebound Policy. Environmental and Resource Economics, Vol. 48, pp. 43-58. Verdolini E., Galeotti M. (2011). At home and abroad: An empirical analysis of innovation and diffusion in energy technologies. Journal of Environmental Economics and Management, Vol. 61(2), pages 119-134. Nesta L., Vona F., Nicolli F. (2014). Environmental policies, competition and innovation in renewable energy. Journal of Environmental Economics and Management, Vol. 67 pp. 369-411. Wang H., Schmidt P. (2002). One-step and two-step estimation of the effects of exogenous variables on technical efficiency levels. Journal of Productivity Analysis, Vol. 18(2), pp. 289–296. Wieczorek Anna J., Marko P. Hekkert (2012). Systemic instruments for systemic innovation problems: A framework for policy makers and innovation scholars. Science and Public Policy, Vol. 39(1), pp. 74-87.

31

ACCEPTED MANUSCRIPT

Tables and figures Table 1 - Share of electrical appliances by origin (own elaboration on COMEXT data)

DK

FR

GR

IT

CE

NL

AC

SE

SI

UK

T

SC

AT

2013 Partner DE IT CZ AT IT FR AT DE SE DE IT DK IT DE ES FR IT DE FR GR DE FR AT IT DE IT FR NL IT DE DK SE DE IT AT SI IT DE FR UK

RI P

Country

DE

MA NU

DE

Weight 0.56 0.21 0.03 0.06 0.23 0.07 0.05 0.40 0.28 0.12 0.11 0.19 0.29 0.22 0.07 0.19 0.39 0.24 0.09 0.03 0.06 0.03 0.02 0.82 0.40 0.17 0.05 0.14 0.28 0.14 0.13 0.10 0.04 0.03 0.01 0.91 0.24 0.17 0.07 0.27

DK

FR

GR

ED

AT

1995 Partner DE IT CH AT IT FR AT DE DE IT SE DK IT DE ES FR IT DE ES GR DE FR ES IT DE IT US NL DE IT DK SE IT DE FR SI IT DE FR UK

PT

Country

IT

NL

SE

SI

UK

32

Weight 0.33 0.13 0.03 0.41 0.19 0.10 0.10 0.17 0.25 0.24 0.09 0.20 0.36 0.23 0.08 0.02 0.38 0.23 0.09 0.02 0.29 0.12 0.08 0.08 0.25 0.14 0.05 0.31 0.30 0.21 0.07 0.05 0.11 0.09 0.02 0.73 0.29 0.27 0.07 0.04

ACCEPTED MANUSCRIPT Table 2 - Baseline estimates Dep variable: log electr consumption

(1)

(3)

(4)

-0.266 -0.278 (-11.50)*** (-11.33)*** [-9.85]*** [-11.58]*** -0.807 -0.739 (-13.45)*** (-11.62)*** [-8.14]*** [-8.46]*** 0.0164 0.0361 (0.10) (0.24) [0.33] [0.67] -0.668 -0.680 (-2.45)** (-2.69)*** [-1.71]* [-2.41]** 0.900 0.901 (8.19)*** (8.45)*** [8.77]*** [7.64]***

-0.237 (-9.21)*** [-7.84]*** -0.641 (-8.58)*** [-8.35]*** -0.264 (-1.63) [-0.52] -0.699 (-2.83)*** [-2.08]** 0.885 (8.28)*** [8.91]***

-0.233 (-9.11)*** [-8.43]*** -0.629 (-8.87)*** [-7.21]*** -0.290 (-1.81)* [-0.58] -0.723 (-3.02)* [-2.20]** 0.887 (8.25)*** [8.34]***

-0.866 (-1.56) [-4.98]***

0.597 (1.05) [0.011] -3.500 (-2.71)*** [-1.18]

-3.208 (-9.36)*** [-12.34]***

3.308 (1.52) [0.25]

0.693 (1.69)* [0.48] -4.241 (-3.53)*** [-1.92]* 3.355 (2.23)** [1.87]* 4.487 (2.18)** [0.62]

MMSLE 0.0201 0.0442 0.456 [3.29]***

MMSLE 0.0165 0.0433 0.381 [3.81]***

MMSLE 0.0173 0.0427 0.405 [3.71]***

T

log(electr_price)

(2)

RI P

log(GDP pc)

SC

log(av. size dwelling)

Residual inclusion from first stage

Variance of the inefficiency component log(Policy index)

ED

log(stock enef patents, total)

MA NU

Share urban population

PT

Share of domestic enef patents over total enef patents

CE

Constant

AC

Model Sigma_u Sigma_v Lambda t-statistics for Lambda (estimated with bootstrap)

MMLWE 0.0556 0.0329 1.689 [1.68]*

Fixed-effects stochastic frontier models (cost inefficiency). Inefficiency is assumed to be half-normally distributed. Student's t statistics in parenthesis. Student’s t statistics estimated with bootstrap (200 replications) in square brackets. * p<0.1, ** p<0.05, *** p<0.01. N=190.

33

ACCEPTED MANUSCRIPT Table 3 - Robustness check (2)

(3)

log(electr_price)

-0.272 (-4.76)***

log(GDP pc)

-0.300 (-3.86)***

-0.236 (-9.22)*** [-6.83]*** -0.655 (-9.79)*** [-7.73]*** -0.289 (-1.77)* [-0.53] -0.734 (-3.08)*** [-1.94]* 0.916 (8.68)*** [8.30]***

-0.263 (-9.26)*** [-7.51]*** -0.557 (-5.73)*** [7.46] -0.342 (-1.58) [-0.95] -0.377 (-1.87)* [-1.37] 0.829 (5.27)*** [7.88]***

T

(1)

RI P

Dep variable: log electr consumption

-0.457 (-1.25)

SC

log(av size dwelling)

-0.617 (-1.43)

Residual inclusion from first stage

MA NU

Share urban population

Variance of the inefficiency component log(Policy index)

0.343 (0.84)

ED

log(stock enef patents, total)

PT

Share of domestic enef patents over total enef patents

Constant

AC

CE

Model Distribution of the inefficiency term Sigma_u Sigma_v Lambda t-statistics for Lambda (estimated with bootstrap)

0.691 0.932 (1.53) (1.20) [0.043] [0.25] -2.560 -4.777 -7.620 (-1.67)* (-4.49)*** (-4.03)*** [-1.63] [-1.88]* 2.442 3.969 5.408 (0.98) (2.25)** (1.77)* [1.18] [1.73]* 2.206 4.876 8.041 (0.92) (2.70)*** (2.25)** [0.34] [0.80] MMSLE MMSLE TRE Half normal Exponential Half normal 0.0413 0.00931 0.0165 0.0441 0.0432 0.0424 0.936 0.216 0.468 [4.13]*** [3.90]*** [4.52]***

True fixed effects stochastic frontier model (cost inefficiency) in column 1 and 2, true . random effect in column 3. Student's t statistics in parenthesis. * p<0.1, ** p<0.05, *** p<0.01. N=190.

Table 4 - Correlation matrix of efficiency scores TFE (1) TFE (4) TRE (3)

TFE (1) 1 0.3402 0.3527

34

TFE (4) 1 0.9874

TRE (3) 1

ACCEPTED MANUSCRIPT Color figures for ENEECO-D-15-00801

DE

GR

IT

DK

0

5

MA NU

10

NL

FR

SE

1995 2000 2005 2010 2015 1995 2000 2005 2010 2015

UK

5

ED

10

SI

0

PT

1995 2000 2005 2010 2015 1995 2000 2005 2010 2015

CE

Graphs by country

AC

Policy indicator

0

SC

5

10

AT

RI P

T

Figure 1 – Policy index trend (own elaboration based on IEA data).

35

ACCEPTED MANUSCRIPT

DE

KR

ES

FR

AT

NL

UK

GE

US

BE

JP

FR

CN

CH

TR

CE

PT

IT

0

20

40

AC

60

US

ED

0

5

MA NU

SC

RI P

10

T

Figure 2 – Patent stock (EPO) in energy efficient technologies for electrical appliances by country.

DE

KR

IT

36

ACCEPTED MANUSCRIPT Figure 3 – Share of 'domestic' patent stock in energy efficient technologies for electrical appliances by country. 2013

0

.2

MA NU

.4

SC

RI P

.6

T

1995

ED

AT DE DK FR GR IT NL SE SI UK

PT

Graphs by year

AT DE DK FR GR IT NL SE SI UK

2013

AT DE DK FR GR IT NL SE SI UK

AT DE DK FR GR IT NL SE SI UK

AC

1995

0

10

20

30

40

CE

Figure 4 – Total, domestic and foreign patent stocks in energy efficient technologies for electrical appliances by country in years 1995 and 2013 (number of patents).

Total enef patents Foreign enef patents

Domestic enef patents

Graphs by year

37

ACCEPTED MANUSCRIPT Figure 5 – Estimated efficiency scores

Estimated efficiency scores Table 2, column 4

DK

FR

GR

NL

SE

SI

UK

AT

2013

True random effect

0

.2

.4

.6

.8

1

Table 4, column 3

DE

DK

FR

GR

IT

NL

SE

SI

UK

2013

AC

CE

PT

1995

ED

AT

RI P

1 .8 .6 .4 .2 0 IT

1995

SC

DE

MA NU

1 .8 .6 .4 .2 0 AT

T

True fixed effect

Table 2, column 1 Estimated efficiency scores

True fixed effect

38

DE

DK

FR

GR

1995

IT

NL

SE 2013

SI

UK

ACCEPTED MANUSCRIPT

Highlights for ENEECO-D-15-00801

SC

RI P

T

“Technology invention and adoption in residential energy consumption. A stochastic frontier approach”

Electricity saving of efficient home appliances is analysed by means of SFA.



Potential endogeneity in the demand frontier function is accounted for.



Technical efficiency is correlated with policy and technology-push mechanisms.



Invention and adoption explain the technical efficiency trend of appliances.



Results stress the importance of supply-push and trade policies to improve EE.

AC

CE

PT

ED

MA NU



39