New Methods For Modeling And Forecasting Of Hidden Periodical Time Series

Cash flow is the survival of company.With the quantitative prediction,we can improve the management efficiency and capital profitability.The statistical analysis of cash flow forecasting has been paid great attention by experts,and its application has a promising prospect.This paper has studied the problem of cash flow forecasting by statistical methods.Because of the periodical character of cash flow,this paper analyzes it with hidden periodical model(HPM),and try to improve HPM's shortcomings.After a lot of exploration and research,such as improving the parameter estimation method,the prediction effect is still unsatisfactory.Then,we break the framework of HPM and propose a new method combining the HPM with machine learning method.Firstly,the HPM is used to preprocess the periodical character of the cash flow.The results of HPM are then used as input feature of random forest model or support vector machine to do prediction.In this paper,the new method is applied to forecast the actual data of many groups.The results show that the prediction effect of new method is better than that of the HPM.The results of this paper are as follows: A new forecasting method based on hidden periodical model and machine learning is proposed,which provides a new technique for the prediction of periodical time series and has important application value.