#research Project
Machine learning, counterfactual prediction, and ex-post assessment of carbon taxes.
Abrell, M. Kosch, and S. Rausch (2019). How Effective Was the UK Carbon Tax? A Machine Learning Approach to Policy Evaluation. Center for Economic Research at ETH Zurich Working Paper No. 19/316.
Abrell, M. Kosch, and S. Rausch (2019). Machine Learning, Counterfactual Prediction, and Policy Evaluation. Unpublished non-technical note.
Program evaluation methods are widely applied in economics and quantitative social science to assess the causal effects of policy interventions based on observational data. Machine learning prediction methods offer new ways of causal inference analysis for policy assessment, including important applications in the domain of climate and energy.
Carbon taxes are commonly seen as a rational policy response to climate change, but little is known about their performance from an ex-post perspective. A new paper by the ENEC group analyzes the emissions and cost impacts of the UK CPS, a carbon tax levied on all fossil-fired power plants. To overcome the problem of a missing control group, we propose a new approach to policy evaluation which leverages economic theory and machine learning techniques for counterfactual prediction.
To avoid dangerous and costly climate change, the disposal space for carbon dioxide (CO2) in the atmosphere is “scarce” and will soon be exhausted (McGlade and Ekins, 2015; IPCC, 2018). In tackling this major 21st-century challenge, and based on an elementary understanding of how today’s market-oriented systems organize economic activity based on scarce resources, economists have long been advocating for carbon pricing as an effective and efficient policy response (Nordhaus, 1994; Goulder and Parry, 2008; Metcalf, 2009). About one quarter of global CO2 emissions are currently regulated under some form of carbon pricing (World Bank, 2018). While a plethora of studies offers ex-ante assessments of carbon pricing using theoretical and quantitative simulation-based work, surprisingly little is known about the ex-post effects of carbon pricing. This, however, is pivotal for designing effective and efficient climate policies in the future.
This paper contributes by providing an ex-post evaluation of a real-world policy experiment of carbon pricing: the UK carbon tax, also known as the Carbon Price Support (CPS). The CPS was introduced to enhance economic incentives for CO2 abatement in the heavily fossil-based UK electricity sector. As the CPS affects the output and operating decisions of all fossil-fueled generation facilities in the UK electricity market, the main challenge arises that no suitable control group or counterfactual exists against which the impact on treated units can be evaluated. In order to estimate the causal effects of the CPS policy intervention, it is thus not possible reated units can be evaluated. In order to estimate the causal effects of the CPS policy intervention, it is thus not possible to use standard program evaluation methods based on comparing treated and untreated units—such as difference-in-differences (DiD), regression discontinuity design, and synthetic control methods (Angrist and Pischke, 2008; Athey and Imbens, 2017). To overcome this problem, we develop and implement a new approach which combines economic theory and machine learning (ML) techniques to establish causal inference of a policy intervention in settings with observational, high-frequency data when no control group exists. We apply our approach to analyze the environmental effectiveness and costs of the UK carbon tax. To our knowledge, this is the first paper in economics to incorporate ML methods to conduct causal inference of carbon pricing.
Our results indicate that in the period 2013-2016 the CPS lowered emissions by 6.2 percent at an average cost of 18 Euro per ton. We find substantial temporal heterogeneity in tax-induced impacts which stems from variation in relative fuel prices. An important implication for climate policy is that a higher carbon tax does not necessarily lead to higher emissions reductions or higher costs.