融合延迟变换和张量分解的金融时序预测算法

Financial timing prediction algorithm fusing delay transform and tensor decomposition

金融时序预测可以为从业人员提供行业变化趋势信息。采用多路延迟嵌入变换将时间序列转化为低秩块Hankel张量,利用Tucker分解将高阶张量投影到压缩核心张量中,对核心张量使用季节性差分自回归滑动平均算法实现对未来的预测。在4个公共数据集上验证了该算法与经典的XGBoost、VAR、SARIMA等算法相比具有更好的计算精度和更少的计算成本。

Financial timing prediction can provide reference information of change trend,thus providing an objective theoretical basis for personnel engaged in financial industry.The time series was transformed into a low-rank block Hankel tensor by multiplex delay embedding transformation,and the high-order tensor was projected into the compressed core tensor by Tucker decomposition.The seasonal differential autoregressive sliding average algorithm was used to predict the future on the core tensor.Four common datasets were used to verify that the proposed algorithm has better computational accuracy and less computational cost compared with classical XGBoost,VAR,SARIMA and other algorithms.