Quantifying the Distributional Impact of Energy Efficiency Measures.

• Author: McCoy, Daire

1. INTRODUCTION

   Residential buildings account for around 30 percent of the United Kingdom's energy consumption and constitute the third largest sector by emissions (BEIS, 2018). With the introduction of recent legislation to achieve net zero emissions by 2050 (UK Climate Change Act, 2019), major steps to reduce emissions occurring in homes will be necessary. Energy efficiency measures will play an increasing role in achieving these reduction targets. In addition to the challenge of the decarbonisation of the domestic energy sector, more than 10 percent of the UK's population still lives in fuel poverty today (BEIS, 2019). For this reason, energy efficiency programmes to date have typically pursued two objectives simultaneously: reducing emissions and combating fuel poverty.

   Yet, policy evaluations often do not take into account the distributional impacts of energy efficiency measures and tend to be based on energy saving estimates from ex-ante engineering models. Previous studies have shown that the engineering estimates of energy savings can be more than three times what is actually realised and the upfront investment costs can be up to twice the actual savings (Allcott and Greenstone, 2017; Fowlie et al., 2018).

   We build on this research tackling the question of how much savings energy efficiency measures actually deliver, a question that is more difficult to answer than it may appear. In contrast to previous research, we further evaluate how these savings are distributed, both across households and over time. Providing answers to these questions is important both for understanding why households appear to under-invest in energy efficiency measures relative to what is socially or even privately optimal, the so-called "energy-efficiency gap" (1) and evaluations of policies aimed at encouraging adoption of energy efficiency measures. Overstated returns make private investments seem more attractive, and makes policies appear more cost-effective than they actually are. This issue is exacerbated by the fact that many policy evaluations rely on ex-ante engineering estimates of savings, rather than observed outcomes.

   Evaluations of energy efficiency improvements tend to take a short time-scale, usually a window of a couple of years on either side of the intervention in order to assess the magnitude of savings. This is despite the fact that time-scale has proven an important factor when examining the impact of building energy codes on energy consumption (Kotchen, 2017), and on the effect of behavioural interventions to reduce energy consumption (Allcott and Rogers, 2014). The impact of energy efficiency measures could vary over time for various reasons. Specific factors related to usage patterns in any particular period may bias results both before and after, while poor installation quality or degradation in the installed equipment may affect the results post-installation. Variation over time could affect the accuracy of measurement, the attractiveness of the investment, or the cost-effectiveness of a government scheme.

   Additionally, it has been shown that rebound effects can exhibit significant heterogeneity across income groups (Aydin et al., 2017). (2) Understanding the way in which cross-sectional and temporal variation interact is crucial to developing a complete picture of the distributional impact of energy efficiency policies and measures. This research extends the above literature by examining both the heterogeneity and persistence of savings associated with installing widely-used energy efficiency measures. In order to conduct this analysis, we exploit an extremely large database of home energy efficiency upgrades and metered energy consumption, (3) covering over four million households and a period of eight years. By combining statistical matching and a range of panel econometric estimators we control for unobserved heterogeneity and selection into various government schemes which funded the upgrades. Another novel feature of this analysis is that our database is a random sample of all households entering into energy efficiency schemes administered by energy suppliers in the UK during this period, thus reducing the potential for "site-selection bias" as identified by Allcott (2015).

   The data allows us to examine the variation in performance depending on when measures were installed, how they perform over time, how this varies by dwelling and socioeconomic characteristics, and ultimately how this affects the cost-effectiveness of measures for different household types. Results indicate significant cross-sectional and temporal variation in energy savings. In particular, the results demonstrate that energy savings are lower for more deprived households. Using predicted income data, heterogeneity becomes even more pronounced showing extremely low savings in the lowest decile and significant differences between the lowest deciles. Quantile regression analysis suggests that differential baseline energy consumption does explain some of the difference between socioeconomic groups, but not all. We also present suggestive evidence that savings diminish over time for loft insulation and heating system replacements. The measures are generally still NPV positive, and compare favourably with the cost-effectiveness of other initiatives, but the returns are much lower than expected.

   It is important to state that more deprived households may trade-off energy savings with increased internal temperatures resulting in increased well-being or health benefits. For this reason we cannot make claims about total welfare gains and how they are distributed. However, the results do raise concerns over distributional factors given how the costs of policies are subsequently levied on households.

   The rest of the paper is organised as follows; Section 2 provides the context in which this analysis takes place; Section 4 the data; Section 3 describes the methodological approach employed and considerations undertaken; Section 5 outlines the results; Section 6 provides the results of robustness checks and sensitivity analysis; Section 7 provides a concluding discussion.

2. BACKGROUND

   The Supplier Obligation (SO), first introduced to the UK in 1994, has become the principal policy instrument for implementing energy efficiency improvements in the domestic sector in the UK (Rosenow, 2012). The Supplier Obligations are an example of a "Tradable White Certificate" (TWC) scheme. These are regulatory mechanisms, employing a market-based approach to deliver energy savings. Theoretically, they can be considered a hybrid subsidy-tax instrument, in which suppliers provide subsidies for energy efficiency upgrades that are then recovered through increased energy prices (Giraudet et al., 2012), having parallels with traditional demand-side management (DSM) programmes in that companies are required to invest in projects that ultimately reduce demand for their product (Sorrell et al., 2009b).

   As outlined in Bertoldi and Rezessy (2008) and Giraudet et al. (2012), SOs have three main features: an obligation is placed on energy companies to achieve a quantified target of energy savings; savings are based on standardised ex-ante calculations; the obligations can be traded with other obligated parties. This flexibility ideally allows suppliers to choose the most cost-effective way to reach their target. Suppliers bear the cost of installations in the first instance, costs are then passed through to their entire population of customers through increases in energy prices (Chawla et al., 2013). (4) Clearly, this may have distributional consequences if certain segments of the population are less likely to avail of the schemes. To alleviate this concern, targets were imposed regarding the proportion of savings to be achieved from lower income groups.

   The former Department of Energy and Climate Change (DECC), (5) sets the savings targets which are then enforced by the energy regulator, the Office of Gas and Electricity Markets (Ofgem). Ofgem sets and administers individual savings targets for each energy supplier. Energy suppliers have various options to achieve their targets such as contracting installers, subsidising energy efficiency products, cooperating with local authorities, delivery agents or supermarkets, or directly working with their customers (Rosenow, 2012).

   Figure 1 gives an overview of SOs from 2002-2012. The first Energy Efficiency Commitment (EEC1) ran from 2002 to 2005, followed by EEC2 in 2005. In 2008, EEC2 was replaced by the Carbon Emissions Reduction Target (CERT) which ran until 2012. In 2009, the Community Energy Saving Programme (CESP) was introduced in parallel with CERT. While the main architecture of SOs did not change, the savings targets and the costs of the delivering the programmes increased over time. Rosenow (2012) provides a comprehensive overview of the main changes in each scheme from 1994-2012 with regards to the target, the costs, social equity implications and other changes in design. The main change concerned the target size, increasing substantially in lifetime savings from 2.7 to 494 terawatt hours (TWh) between 1994 and 2012 (Rosenow, 2012).

   From 2002, all programmes included a target for disadvantaged households and fuel poverty increasingly came to the fore. Eventually, CESP only allowed projects to be carried out in specific low income areas of Britain, the lowest 10-15% of areas ranked in Income Domain of the Indices of Multiple Deprivation (Hough and Page, 2015). Thus, CESP was only available in certain geographical regions. Furthermore, CESP introduced a new bonus structure that incentivised the installation of multiple measures in a single dwelling and the treatment of as many dwellings as possible in the same area (Duffy, 2013). Table 1 summarises the key features of the schemes under consideration.

   A key feature of all previous evaluations of the above policies is that the energy savings achieved...