基于ARMA模型和VAR模型的中国债券市场

选取上海证券交易所企业债和国债月度数据，利用遗传算法对静态利率期限结构NSM参数模型进行求解，进而拟合较为精确的企业债和国债的利率期限结构，据此计算出企业债的信用价差。数据一部分作为样本内拟合区间，另外一部分作为样本外预测区间以检查模型的预测精度。通过建立自ARMA样本外预测模型和VAR样本外预测模型分别对我国债券市场信用价差进行预测，最后比较两种模型的预测精度。结果表明VAR模型对于信用价差短期预测较为准确，而ARMA模型对于较长期预测较为准确。

Comparison of credit spread forecasting based on ARMA and VAR models

In order to study China’s bond market, this paper selects the monthly transaction data of the Shanghai Stock Exchange and by using the NSM model, combined with a genetic algorithm, the interest rate term structures of government bonds and corporate bonds are obtained along with the credit spread which is the difference between the two interest rate term structures. Part of the data was taken as the fitting sample, and the remaining part of the data was taken as an out-forecasting sample to check the accuracy of predictions made using the model. The ARMA and VAR out-forecasting models have been established in order to forecast China's bond market credit spreads, and the precisions of the two models have been compared. The results show that the VAR model is more accurate for short-term forecasts, and the ARMA model is more accurate for longer-term forecasts.