Research Of Shanghai Composite Index On Time Series Analysis

Prediction of stock price has become more and more popular, howeverthe stock price time series itself has a complexity, diversity and gooddegeneration, and there are many factors affecting the stock changes due tothese factors affecting the stock market, some can be measured, some cannot,it is difficult to calculate all the factors are quantified by calculation andevaluation in order to evaluate scientific research. Modern statistics for thestudy of problems in the stock market is often valued fitting effect within thesample, optimization of various models to achieve a perfect sample returnwithin the data, while the fitting precision of the sample tends to make theincreasingly stringent requirements people ignore the robustness of theresearch model, not the excellent fit of the sample is extended to samples,mutations in the stock market, especially in the face of emerging long-termtrend, does not have the ability to respond quicklyThe time series models can be divided into fundamental analysis,technical analysis and statistical analysis, statistical analysis also can be divided into two categories: traditional statistical methods and computationalintelligence methods. This paper reviews and summarizes the two classmethod to date on current stock price time series prediction, and thus basedon existing domestic and international research on the status of the stockprice time series prediction of existing research conducted commentary,pointing out the current study problems and stock data since1998empiricalanalysis, the state space model and artificial neural network models wererolling forecasts empirical analysis of two rolling and rolling forecastsupdated model under a fixed model for model the goodness of fit, volatilityand long-term stability. Comparing four models, it shows that the goodnessof fit of the simple models though poor in the short term, but in the long term,especially when facing market mutated, performance better than complexmodel. For complex models, because of overfitting within the sample,short-term forecasting always volatile, fit of goodness for long-term forecastsoften appear unstable. These conclusions provide some reference value toselect an empirical model for the future.