Prediction Of Stock Index Differentiation Based On Time Series Model And Semi-parametric Model

In the process of China's economic development,financial market plays the role of a reservoir,which can not only store capital but also convert saving into investment flexibly and quickly.It promotes the rational distribution of capital in the whole society The stock market,an important and special branch of the financial market,has a mutual influence on macro economy.It is significant to research the development of stock market for analyzing macro economy.Compared with a single index,the trading pattern which refers to the differentiation of two stock indexes is gradually popular in the stock market,because it has many advantages such as obvious trend,expanded fund management quota,less risk.This thesis calculates the differentiation value and its daily range based on closing prices of SSE 50 and CSI 500 from 2016-11-18 to 2019-11-18.Build and analyze mathematical models of differentiation values' daily ranges to forecast,study the influencing factors in depth and make suggestionsResearching of stock market usually bases on time series model.After analyzing the feasibility,the thesis establishes traditional ARIMA model to analyze the linear correlation between current and past values and predict five future values.On the basis of the ARIMA model,establish partial linear model to forecast residuals by adding five influencing factors,namely the sum of differentiation's band amplitudes which are greater than threshold,range of CSI 500,range of SSE 50,minimum value of differentiation and range in the first hour.The predicted values of differentiation values'daily ranges are obtained again by the combination of ARIMA model and partial linear model.In view of the scientific nature of the research,three sets of models are established in this thesis and the sample size is gradually increased.Each set of models consists of an ARIMA model and a composite model,and the sample sets of two models in the group are sameBy comparing the prediction results of two types of models,it is found that the accuracy of the combined model is 20%higher than that of ARIMA model,indicating that the combined model has significant correction effect on the prediction of time series model.In addition,the prediction grades of time series model are single and result of long-term prediction is highly correlated with that of short-term prediction,so the flexibility of model is poor.After the correction of partial linear model,the flexibility of model is enhanced.What's more,combined model has more stable prediction effect and higher reliability.Comparing three sets of models,the prediction effect will be better and stabilize with the increase of sample data.Besides showing the linear relationship within its own sequence,the combined models also indicate the influences of various factors on the differentiation values' daily ranges.Generally speaking,it is more effective to establish combined model for the research object of this thesis.According to the prediction results of three sets of models,put forward the following reasonable suggestions:First,adjust the number of trade according to differentiation values' daily range during the process of one-day trading.A large predicted value represents a larger profit and a more stable trend,at which time the number of transactions can be appropriately increased.Second,the results of ARIMA model show that the current value has linear correlation with sequence values of past three or four days,so paying more attention to the trend of previous days is helpful to understand the differentiation of that day in advance.Third,bands with a range of 0.5 or more are considered to be highly profitable and usually occur once or twice a day.They have obvious positive influence on the daily range of differentiation.In the process of their formation,late trading can both guarantee profits and reduce risk.Forth,in the expressions of three combined models,the coefficients of range in the first hour are both very small.Some of them are bigger than zero and some are smaller.This shows that the shock of first hour doesn't have a clear impact on the all-day situation,so it is not rational to predict the differentiation of the whole day based on the early shockFifth,the range of CSI 500 has a positive effect on the dependent variable.And it has a stronger impact than the range of SSE 50.Therefore,more attention should be paid to the differentiation dominated by CSI 500.Sixth,the coefficient of minimum differentiation value is negative in the combined model.The smaller the minimum value,the greater daily range.So pay attention to the large increase from bottom when the differentiation value reaches a small value.