Tradable performance standard (TPS) is an important policy instrument for mitigating CO2 emissions in developing countries, who play an essential role in achieving drastic global carbon emissions reduction. Whether TPS effectively reduces firm-level emissions in the developing country context remains an unknown empirical question. This paper takes the first step in answering this question based on a policy experiment in China. Since 2013, China has introduced carbon emissions trading systems in eight regions based on TPS (ETS pilots). This study provides the first ex-post evaluation of the effects of these ETS pilots on emissions from coal-fired thermal power generation facilities using staggered and dynamic difference-in-differences models. This study uses a novel dataset from NASA’s Aura satellite to measure SO2 emissions at the facility level. Contrary to common belief, results show that although SO2 emissions of all facilities declined steadily from 2010 to 2019, SO2 emissions of facilities covered by the ETS pilots (ETS facilities) increased by about 5-7% relative to those of non-ETS facilities. Moreover, the relative increase in SO2 emissions of ETS facilities grew over time. A model is developed to show that the implicit output subsidy from the TPS design could increase the output of cleaner facilities, leading to more SO2 emissions.