Factor Models For High-dimensional Time Series Forecasting

High-dimensional time series prediction is widely used in many scenarios,such as monitoring and alarm,stock analysis,marketing prediction and so on.In many cases,decision makers are required to grasp the characteristics of massive time series data in time and make fast decisions.In analysis,the main consideration is the extraction method of dynamic factor model,in the purpose of the dimension reduction at the same time,people can predict time series through the common factor.Existing methods,there are problems about the high-dimensional time series data structure information and the absence of observable factors of temporal information.In this paper,a factor model based on high-dimensional time series is proposed.This method takes into account the dependence of observable factors on the time series,and can also quickly reduce the dimension of unobservable factors without destroying the time relationship.The linear relationship matrices of the row loading matrix,the column loading matrix and the observable matrix are estimated,and the identifiability of the autoregressive data matrix is proved.Finally,this paper constructed a high-dimensional time series matrix using the reservation rate of 11 flights for 365 days in 2018,and analyzed the high-dimensional dynamic network through projection estimation,matrix least square method and other methods.By comparing different dimensions,it is found that the error rate of the proposed method is less than 0.1 and the speed is faster.