In many industries, the cost of abating emissions depends on firms' ability and incentives to switch to cleaner fuels. I develop a dynamic production model that recovers heterogeneous marginal abatement costs from plant-level production and fuel-use decisions. Due to historical investments, plants differ in how productively they can use different fuels. They also differ in the set of fuels they use, but can pay fixed costs to change their fuel sets. I estimate the model using detailed panel data from Indian steel plants, leveraging minimal assumptions about optimal input choice and the technology frontier. Heterogeneity in fuel productivity creates technology lock-in within plants but flexibility across plants. Implementing a carbon tax that reduces emissions by 50 percent would reduce producer and consumer surplus by 4%, compared with 7% when fuel-specific productivity heterogeneity is ignored. The lower cost of reducing emissions arises because plants with a comparative advantage at using cleaner fuels are less exposed to the tax and become relatively more competitive, reallocating output away from coal-intensive producers. Output reallocation accounts for 60% of the aggregate emissions decline, whereas within-plant fuel switching accounts for only 15% of emissions reductions under carbon pricing. 

[Code] [Online Appendix] [Slides]

This paper examines how plants adjust their production in response to asymmetric carbon pricing. When plants compete across areas, asymmetric regulation can lead to carbon leakage, shifting emissions from regulated to unregulated areas. I build a production model with multiple fuel inputs, imperfect competition, and region-specific carbon taxes. Using publicly available Canadian plant-level data on a wide range of air pollutants, I invert the chemical reactions from combustion to back out plants’ fuel usage. I then estimate the model by exploiting variation in the British Columbia (B.C.) and Quebec carbon taxes, which were implemented in 2008 and 2007, respectively. Findings indicate substantial emissions reductions in British Columbia, with 95 % confidence intervals ranging from 7 % to 48 %, and no reduction in Quebec. Contrary to theoretical predictions of carbon leakage, the analysis reveals no statistically significant shift in production toward unregulated provinces. A detailed decomposition reveals that the absence of leakage was primarily due to the regulated plants’ ability to absorb the tax by switching from oil to natural gas and by reallocating output from dirtier to cleaner plants within British Columbia.