Abstract: The price of carbon assets fluctuates heavily because of the global economy, politics, energy, and so on, thus it has been of realistic significance to have research on the risk measurement of carbon market. In this paper, EUA and CER markets are taken as the research objects, and the performance of CAViaR model and GARCH-GED model in measuring the risk of carbon markets under the different prediction intervals and confidence levels are compared, finding that:(1) CAViaR model is better than GARCH-GED model in fitting and prediction; (2) CER market has greater uncertainty relative to EUA market; (3) when predicting 1%VaR, the CAViaR model is instable. In hope of a better prediction effect, this paper takes the combination of CAViaR model and EVT is taken to predict 1%VaR, finding that the prediction of EVT-CAViaR model is more steady and reliable under the high-risk prediction intervals and the CER market, therefore a conclusion can be made that this new method promises to partly improve the prediction accuracy of the extreme risk of carbon markets.

Key words: carbon market, CAViaR, EVT, VaR