# 余额宝资金流量预测方法研究

1.    根据时间序列的散点图、自相关函数和偏自相关函数图识别其平稳性（仅取平稳序列作为历史数据）。
2.    根据所识别的特征，建立相应的时间序列模型。经过平稳化处理，逐步建立一次，二次，三次和四次多项式回归模型。
3.    利用已通过检验的模型对未来一个月每天的申购赎回情况进行预测。

ABSTRACT
In recent years, the rapid development of internet finance, the birth of many new things, such as Alibaba Group Ant Gold clothing launched the balance treasure. Since the birth of June 13, 2013, the balance has a huge user base, hundreds of billions of dollars a day of capital flow, the flow of the number of projects and amounts are extremely large. Therefore, we need to adopt various methods and models, according to the financial inflow and outflow of many balance users over the past period, a relatively accurate forecast of the daily flow of funds can be used to ensure that the funds available are invested in other areas in order to maximize revenue in the absence of a crisis of financial flows.Regression analysis (Regression) is a computational method and theory that studies the specific dependencies of a variable (interpreted variable) on another (some) variable (explanatory variable). Starting from a set of sample data, the mathematical relation between variables is determined to carry out various statistical tests on the reliability of these relationships, and to find out which variables have significant impact and which are not significant from the variables that affect a given variable. The desired relationship is used to predict or control the value of another specific variable based on the value of one or more variables, and the accuracy of the prediction or control is given.The main regression fitting methods include smoothing curve fitting, least squares interpolation, polynomial fitting, trigonometric fitting, Gauss curve, and so on, based on the prediction precision and the difficulty of realization, we use polynomial fitting to establish polynomial regression model.
Polynomial regression models (Polynomial Regression model, PNRM), which is one of the classical prediction methods in time series analysis. The specific steps of the project forecast are as follows:
1. According to the scatter graph, autocorrelation function and partial autocorrelation function graph of time series, the smoothness is identified (only stationary sequence is taken as historical data).
2. The corresponding time series model is established based on the identified features. After the stabilization process, a polynomial regression model was set up, two times, three times and four times.
3. Use the tested model to forecast the daily redemption of the next one months.
Calculates the error of daily purchase and redemption Total and actual value of all users on test set, the smaller the error, the higher the forecast accuracy. In order to reduce the error, the future improvement direction is to use the ARIMA model for modeling and forecasting.
Key Words: capital flow; time series analysis; PNRM model;ARIMA; autocorrelation function

1.1 研究目的及意义    6
1.2 国内外研究现状    6
1.3 研究内容和目标    8

2.1 数据预处理    9
2.1.1 原始时间序列    9
2.1.2 平稳性检验    10
2.2 拟采用的技术方案    10
2.3 分析与改进    12

3.1 模型的建立    14
3.1.1 数据的分析    14
3.1.2 多项式回归模型    16
3.2 模型的求解    16
3.2.1 平稳时间序列    16
3.2.2 数值计算求解    18

4.1 总结    32
4.1.1 思路概括    32
4.1.2 主要创新点    32
4.1.3 局限和不足    33
4.2 改进方向    33