Direct and Indirect Energy Rebound Effects in German Households: A Linearized Almost Ideal Demand System Approach.

• Author: Schmitz, Hendrik

1. INTRODUCTION

   Improvements in energy efficiency are often considered to be effective and cost-efficient for mitigating both energy consumption and greenhouse gas (GHG) emissions. However, these advances are prone to triggering rebound effects--behavioral changes that stem from the decreased price of an energy service due to the increased efficiency, leading to an increased demand for that specific service. Direct rebound effects are a standard reaction to a price decrease and are easily explained by basic microeconomics. They consist of an income effect and a substitution effect. Additionally, there exist indirect rebound effects. These occur when an efficiency increase related to one good or service increases the demand for other goods and services. For example, buying a more fuel-efficient car can cause a number of different reactions: driving more frequently and longer distances in the new car is a direct energy rebound effect, whereas using some or all of the monetary savings to undertake more trips by airplane would be an indirect effect.

   The magnitude of rebound effects has implications for energy policy: namely, the higher a rebound effect is, the less effective efficiency improvements are in achieving environmental goals, because more of the potential savings remain unrealized due to higher consumption. Since these improvements are sometimes induced by government activity, for example in the form of mandatory efficiency standards or research funding for efficiency technologies, rebound effects should be taken into account when evaluating these policy measures ex ante and ex post. Conversely, due to the close relationship between price elasticities and rebound, high rebound effects suggest that households are highly price-sensitive with regard to the demand for energy. High price sensitivity increases the effectiveness of price-based measures, such as taxes on energy consumption.

   Ultimately, one of the main goals of energy and environmental policy is not only to reduce energy consumption but also to decrease GHG emissions. Most goods and services have some amount of embodied emissions that are released during the production process, even though the exact relationship between consumption and energy use is highly heterogeneous across goods. For some goods, this emission might be largely invisible to the consumer, such as sleeping in a hotel or going to the movies. Other goods, such as furniture or books, typically require no energy while in use, but still need energy along their production chains. In order to measure the real overall effects of an efficiency improvement, we need to include its impact on the demand for all goods and services of an economy, not just energy goods and services.

   Numerous studies have estimated rebound effects across different goods and services and for different countries. (1) Many of those exploit the close relationship between price elasticities of energy demand and rebound effects. However, this identity relies on strong assumptions. Namely, energy prices need to be exogenous, energy efficiency needs to be constant, and energy service demand needs to be dependent on energy service prices only (Sorrell and Dimitropoulos, 2008; Chan and Gillingham, 2015). Since these assumptions are unlikely to hold in practice, Hunt and Ryan (2014) suggest explicitly incorporating energy efficiency in the estimations so that price elasticities can serve as unbiased estimates of a rebound effect. However, most studies so far have not addressed this issue. Furthermore, most empirical estimations of direct and indirect rebound effects only take into account immediate energy use, omitting the embodied energy that is used in the production and shipping of all goods, including food or clothing (Chitnis and Sorrell, 2015). In view of this gap in the current literature, our research question is the following: what is the magnitude of direct and indirect rebound effects for energy carriers used by German households? To the best of our knowledge, our study is the first to include the embodied energy and C[O.sub.2] intensity of all goods and services consumed by households, while explicitly accounting for energy efficiency in the estimations. The analysis is undertaken for the case of Germany.

   Several studies have estimated direct and indirect energy rebound effects for different countries and time periods by using a systems approach and by incorporating both income and substitution effects. Brannlund et al. (2007) investigate monthly household data for Sweden for 1980-1997. Employing a three-stage budgeting model that separates spending into durable and non-durable goods before using four groups in the second stage, and 13 goods and services in the third stage, the authors estimate own- and cross-price elasticities in a Linearized Almost Ideal Demand System (LAIDS) framework. They find excessively large total rebound effects, ranging from 121% to 175% across categories. (2)

   Mizobuchi (2008) examines monthly Japanese expenditure data ranging from 1990 to 1998. Unlike Brannlund et al. (2007), who simulate a costless efficiency improvement, Mizobuchi (2008) explicitly takes capital costs into account. To this end, the author considers actual efficiency and price data for different energy-consuming appliances such as TV sets and water heaters. His paper finds a combined rebound effect of 27% when including capital costs and 115% when excluding them, the latter result being similar to the lower end of the values found by Brannlund et al. (2007).

   Chitnis and Sorrell (2015) estimate direct and indirect rebound effects for a wide variety of goods and services consumed by private households in the UK, encompassing both durables and non-durables. By mapping GHG emissions to the different categories, the authors manage to evaluate the full impact of rebound effects for households in the UK. Their paper finds a combined rebound effect for gas, electricity, and vehicle fuels of 55%, with most of this effect stemming from substitution rather than from income effects. Direct effects are significantly larger than indirect ones. The authors find that the direct rebound is mostly driven by substitution effects, while indirect rebound is caused by income effects. Due to the limitations of their approach, the authors suspect that their results overestimate the magnitude of the combined rebound effect. The present paper conducts a similar analysis to that of Chitnis and Sorrell (2015) but using a German dataset. It also extends the analysis by representing energy efficiency improvements through the incorporation of time trends, price growth rates, and price asymmetries across various model specifications, respectively.

   None of these three studies incorporate efficiency explicitly in their estimations, as proposed by Hunt and Ryan (2014). So far, the only contribution that does so is that of Peters and McWhinnie (2015). Here, the authors use micro-level data from a repeated cross-sectional household survey for Australia. The rich demographic nature of the dataset allows them to estimate a variety of models that find robust direct rebound effects. These effects range from 103% to 120% for electricity and from 138% to 284% for natural gas. However, the authors' analysis is limited to finding direct rebound effects, and it disregards any kind of indirect rebound stemming from either other energy sources or other goods that contain embodied energy. There are several other studies that estimate direct and indirect rebound without taking substitution effects into account, focusing only on income effects instead (e.g. Druckman et al., 2011; Thomas and Azevedo, 2013). If substantial substitution effects exist, focusing only on income effects would significantly bias rebound estimates. For a recent overview of these studies, we refer to Chitnis and Sorrell (2015).

   As Hunt and Ryan (2014) show, estimating rebound effects by using price elasticities can lead to biased estimates when energy efficiency is not explicitly accounted for in the estimations. Given that the exact energy efficiency for most energy services is usually unobserved, it remains an open question how to implement efficiency in the empirical estimations (Hunt and Ryan, 2014). A straightforward option is to implement a linear or non-linear time trend. However, this assumes that all efficiency increases are exogenous. Since increasing efficiency might also be induced by rising prices, another way is to include past prices for different energy sources, for example by using price growth rates or relative prices. In the model specifications used, we implement (a) a linear and quadratic time trend, (b) time trends and lagged price growth rates, (c) time trends and lagged cumulated price increases and decreases reflecting asymmetric price responses, in order to overcome the shortcomings of previous research.

   The remainder of this paper has the following structure. In Section 2, we describe our methodological approach and the data used. Section 3 presents and discusses the results of our econometric estimations. Section 4 presents a conclusion and suggests avenues for future research.

2. DATA AND MODEL

   2.1 Methodology

   We derive direct and indirect rebound effects for all goods and services consumed by private households in the German economy by using a LAIDS approach. The original AIDS was first proposed by Deaton and Muellbauer (1980) and possesses a number of desirable characteristics that contribute to its long-lasting popularity and widespread use in analyzing consumer demand. Specifically, the AIDS allows perfect aggregation across consumers and can serve as a first-order approximation for any demand system (Holt and Goodwin, 2009).

   The original AIDS model is defined by Deaton and Muellbauer (1980) as

   [w.sub.i] = [[alpha].sub.i] +[summation over (j)][[gamma].sub.ij]ln([p.sub.j]) + [[beta].sub.i]ln(x/P)+[[epsilon].sub.i] (1)

   where i, j [member of]...