A Comparative Study Of Two Forecasting Models Of Capital Inflow And Outflow

With China’s rapid economic growth,a large number of financial-related companies are rising rapidly,such as Ant Financial Services Group’s Yu’e Bao.Hundreds of millions of users have caused Yu’e Bao to have a lot of money flowing in and out every day,and the pressure on capital management is also increasing as the number of users expands.Under the condition that the risk of capital flow should be reduced and the normal operation of daily business should be made,it is extremely important to make an accurate prediction of the future capital flow.As can be seen from relevant literature and information,the models commonly used today are the BP neural network model and the ARIMA time series model.Because the order of magnitude of the raw data is too large,it needs to be processed into smaller order of magnitude sequence data before use.The BP neural network here uses one-step iterative method to predict one result at a time,and then combines the prediction results with the previous data to enter the next prediction again,according to this cycle to get all the predicted results.Before forecasting,the model needs to be trained,the quality of training depends on setting the parameters of the model,as long as the training model converges,that means once the error in the training process reaches the pre-set error,the model stops training.And the trained model can be used for the next prediction.In order to compare the fitting effect of the network model,the time series model is used to analyze it.The original data is unstable time series data,so it is necessary to transform it into a stable time series,and then make white noise test and normality test to ensure that the data can use ARIMA model.Then the stable non-white noise sequences are identified and verified,the possible models are screened by AIC criterion,and the models with the lowest AIC value are selected to predict the inflow and outflow of funds.The results show that the above two models have higher fitting accuracy,and the prediction effect of the time series model is better than that of the neural network model.From the prediction results,the two models are suitable for forecasting funds,but each has its own advantages and limitations.Therefore,we need to fully understand and analyze the original data and then choose the appropriate model in future research,at the same time we need to try different models and algorithms.