A statistical evaluation of UK energy policy: The climate change levy and climate change agreements

The Climate Change Levy (CCL) is an energy tax introduced by the government in 2001. Firms in the industrial, agricultural, commercial, and public sector are levied on each kw/h of energy consumed; tax rates vary between electricity, gas, and solid fuels. Fearing a loss of competitiveness for energy intensive businesses, the Climate Change Agreements (CCA) scheme was designed to lighten the burden of CCL. The current iteration, phase 2, began in April 2013 and reduces the CCL by 90% for electricity and 65% for all other fuels in exchange for meeting negotiated emissions targets.

The literature studying CCA and CCL focuses on ex-ante evaluations of the policies. Limitations in data that prevented ex-post evaluation of the policy are diminishing due to advances in computer science. Using confidential administrative data provided by the Department for Business, Energy, Industry, and Strategy (BEIS) this PhD thesis aims to study the effects of the policies through modern econometric and machine learning methods. The core investigation will aim to answer the following questions:
1. What was the direct impact of the CCA on energy demand?

2. How did firms achieve the required reductions in energy consumption for schema participation?

3. What kind of characteristics were common among firms that strongly reacted to CCA? What kind of firms ceased operations?

The results of the thesis can be used by policy makers to inform forward-going energy policy.
The paper currently in progress answers the first question and studies the effect of the CCA on observed energy consumption in policy participants. Using the novel Changes-in-Changes estimator developed in Athey and Imbens, 2006, it becomes possible to investigate heterogeneous impacts of the policy across the distribution of outcomes. In essence, one can see how each firm reacted to participating in CCA, regardless of size. Schema participant and electricity data were provided by BEIS, however, the data was not ready to use due to not having shared uniquely identifying variables. A substantial amount of time was spent to devise a matching algorithm that utilizes "edit distance" metrics, i.e. algorithms that measure word similarity, to match addresses of facilities and electricity meters in the scheme. Preliminary results indicate that small facilities increased energy demand while it decreased for large facilities. The estimates combined with elasticity calculations imply that emissions targets were tight and difficult to satisfy for large facilities. The results of this study are submitted for presentation at the Energy Evaluation Europe 2020 conference, part of the CCA evaluation conducted by BEIS, and part of my upcoming upgrade seminar.
The second question will be answered using the innovative causal forest method developed in Wager and Athey, 2018, which builds on random forest algorithms to estimate heterogeneous treatment effects. In this application, the aim is to study how firms satisfied the emissions targets. Using data from the Annual Business Survey (ABS), which contains variables such as turnover, energy expenditure, and employment, it becomes possible to estimate whether firms raised efficiency or decreased output to meet targets. To gain access to the ABS data hosted by the Office of National Statistics, I obtained the required accredited researcher certification.
The last question will be answered by using a support-vector machine, a classification algorithm, to group data in the ABS depending on their responsiveness to CCA estimated in the first question. Ultimately, the goal is to understand what kind of firm metrics, such as investment in capital/technology or time-trends in energy efficiency, strongly predict the response to the policies. In a similar vein, the methodology will be used to understand what type of firms in CCA and CCL experienced facility closure post policy.