As an important tool to address climate change, the carbon market is an important path to achieve the “double carbon” goal in China by helping to reduce greenhouse gas emissions through financial means. However, due to the short development history of carbon finance market in China, the research on risk management and empirical evidence of carbon finance market still needs to be further deepened. We use the new dynamic semi-parametric model model to fit and predict the risk of each carbon market, and compare the performance of traditional models in measuring carbon market risk under different VaR predictions. The results show that, from the data point of view, China’s carbon financial markets are severely fragmented, with huge differences in transaction prices and volumes among markets, and different degrees and frequencies of price fluctuations in different markets; in terms of model fit and model prediction, the dynamic semi-parametric model performs better than other models overall; meanwhile, this paper also focuses on the effect of extreme value theory on the CAViaR model. It is found that the EVT-CAViaR model has an improvement effect on the areas with extreme fluctuations.