Does Free Allocation Slow Decarbonization ? Installation-Level Evidence from the EU ETS, Phases III and IV

Farhad Zeynalli 

Installations at high risk of carbon leakage receive free allowances from the European Union Emissions Trading System. Free allowances are allocated to prevent energy-intensive factories from relocating outside the EU. However, whether free allocation slows decarbonization remains an open question. In this paper, I test this question directly.
The panel was built at the installation-level. The data were obtained from the European Union Transaction Log (EUTL). In total, 112,787 installation-years from 2013 to 2023 were observed. To avoid the mechanical bias in derived variables such as surplus (total free allocations-verified emissions) and generosity ratio (total free allocations/verified emissions), free allocation was scaled by fixed base-year (2013) emissions and then lagged by one year. Log verified emissions was then regressed on the scaled variable and Two-Way Fixed Effects (TWFE) estimator with installation and year fixed effects was used.
The estimated effect is very small, close to zero. The coefficient is 0.0006 with a p-value of 0.78. A unit increase in scaled allocation changes emissions by 0.06%, and this is statistically insignificant. This indicates that large free allocations did not slow installation-level decarbonization in Phases III and IV.
The result remains true even without winsorization and when an emissions floor is added. It also holds when removing the 2020 pandemic year. Changing the outcome to emissions intensity using a Eurostat production index does not change the findings either. A difference-in-differences (DiD) test around the 2021 allocation cut gives a large negative coefficient. However, that estimate is invalidated by pre-trends. The Callaway & Sant’Anna, 2021 estimator finds no effect when installations losing most of their free allocation are compared with those losing almost none. Splitting that comparison by sector suggests the cut is associated with lower emissions only in the energy-intensive (carbon-leakage) manufacturing sectors. There are no detectable effect elsewhere. This localized pattern is suggestive rather than confirmed.