基于VAR模型的多元时间序列预测研究

Research on multivariate time series prediction based on VAR model

为了研究因果序列使用VAR模型对多元时间序列预测的影响,选择2014年1月1日至2014年4年30日的中国银行数据的五组时间序列,即开盘价、最高价、最低价、关盘价和成交量进行多元预测,前期进行格兰杰因果性检验发现前四组序列存在因果关系,成交量序列与前四组关系不大。基于此,实验提出采用VAR+ARIMA模型和纯VAR模型对中国银行数据进行预测。结果表明:因果序列采用VAR模型预测效果,比非因果序列预测效果更佳。并且非因果序列预测会影响原因果序列预测的拟合效果。

In order to explore the influence of VAR model on multivariate time series prediction,five sets of time series of Bank of China data from January 1,2014 to April 30,2014,namely,open,high,low,close and volume,are selected for multivariate prediction.Through Granger causality test,it is found that there is a causal relationship between the first four groups of sequences,and the volume has little relationship with the first four groups.Based on this experiment,VAR+ARIMA model and pure VAR model are proposed to predict the data of Bank of China.The results show that the causal sequence uses the VAR model to predict the effect,which is better than the non-causal sequence prediction.In addition,non-causal sequence prediction will affect the fitting effect of causal sequence prediction.