Research on Modeling Method of Capital Inflow and Outflow of Internet User

• 1、Technology Management Office of Traffic Management Bureau of Public Security Bureau, Wuhan 430030
• 2、Computer Science and Technology School, Wuhan University of Technology, Wuhan 430070

Abstract：In recent years, China\'s Internet finance has developed rapidly, and the daily business of Internet wealth management companies involves a considerable amount of capital inflows and outflows. In order to reduce the liquidity risk of funds, it is necessary to predict the inflow and outflow of funds to help them conduct efficient operations and risk management. Based on the official historical data provided by Ant Financial, this thesis forecasts the daily total inflow and outflow of funds in August 2014. The training set is selected by using the observation of line graph and the DF detection method, and then the prediction model is established. There are four kinds of prediction methods: 1. Directly prediction by using BP neural network; 2. Directly prediction by using RNN; 3. After the original data is decomposed by STL, the BP neural network is used to predict the trend, and finally the data to be predicted is calculated; 4. After the original data is decomposed by STL, the RNN is used to predict the trend, and finally the data to be predicted is calculated. The experimental results show that the method of combining STL decomposition and neural network is more accurate.

Keywords： total capital inflows and outflows time series stationarity analysis STL decomposition neural networks