基于LSTM-EGARCH组合模型的VaR碳交易风险度量研究
Research on VaR Carbon Trading Risk Measurement Based on LSTM-EGARCH Combined Modeling

基于碳交易价格收益率信息，从深度学习理论视角研究碳交易市场系统风险，构造了LSTM-EGARCH波动率动态预测模型，分别采用EVT半参数方法，正态分布、T分布参数方法估计标准化收益率分位数，建立LSTM-EGARCH-VaR风险度量模型。模型对比传统EGARCH-VaR模型，克服了波动率变化的线性假设和残差序列的独立同分布假设，其基于风险预测失败的LR检验结果表现，其风险度量结果的准确率均提升38%以上，显示出了深度学习理论在预测领域的优势。
Based on the information of carbon trading price returns, this study investigates the risk of the carbon trading market system from the perspective of deep learning theory. A LSTM-EGARCH volatility dynamic prediction model is constructed, and EVT semi-parametric method, as well as normal distribution and t-distribution parameter methods, are employed to estimate the quantiles of standardized returns. A LSTM-EGARCH-VaR risk measurement model is established. Compared with the traditional EGARCH-VaR model, this model overcomes the linear assumption of volatility changes and the in-dependent and identically distributed assumption of residual sequences. The LR test results based on risk prediction failure show that the accuracy of risk measurement results is significantly improved, demonstrating the advantages of deep learning theory in the field of forecasting.