Weather extremes caused by air pollutant emissions that the carbon dioxide represents have been a threat to the existence and development of human beings. It has become one of the important ways for coping with weather extremes to establish marketable carbon emission reduction mechanism. One of basic functions of carbon market helps with carbon emission reduction by carbon emissions trading and the key of market’s function efficiency is that VaR can be measured accurately. Therefore, it is of realistic significance to give research on the risk measurement of carbon market.Traditional econometric models tend to get inaccurate results for the risk prediction of carbon market because of the limitations of functions themselves although they can work based on the risk characteristics. For this reason, this paper takes neural network quantile regression (QRNN) model to predict risk, which promises to get better results. This paper takes the consecutive contracts trading prices of EUA and CER futures from March 17,2008 to August 28,2013 as the data sample. First of all, this paper builds QRNN model to respectively predict the market risk under the normal and extreme circumstances, which is compared with the widely used GARCH-GED model and CAViaR model. The results of empirical analysis show that:(1) QRNN model performes best when prediting the risk under the normal range. QRNN model is better than GARCH-GED model and CAViaR model in the prediction of 5% VaR; (2) However, all the models heavily underpredict the risk under the extreme range. Considering that EVT model can works well in describing the characteristics of extreme risk and in order to get the relatively accurate prediction of the extreme risk, furthermore, this paper builds the QRNN-EVT model to have a research on 1%VaR, finding that QRNN-EVT model can greatly improve the accuracy of prediction. |