All-pass Models And Traditional Time Series Models In The Chinese Stock Market In Empirical Research

The time series technology is an important tool in study of stock market. This paper introduces traditional time series models and a new model called all-pass model as well. Two methods are considered in solving the all-pass model parameters. After study, we come to some valuable conclusions.These are the primary contents of this paper:First, we observe that the stock returns of our country are non-Gaussian distribution and have volatility clustering when we study some basic statistics indexes.Second, traditional time series models are used to analyze China stock market. Models combined GARCH and ARCH with ARMA respectively are fitted stock returns. They can mimic the volatility behavior. The study results are fitted together with the former conclusions.Third, this paper introduces some basic theory of all-pass model and the least absolute deviation criterion. The parameter scopes of causal all-pass model are deduced by Jury criterion.Fourth, Non-Linear Programming is adopted to solve the LAD estimator. In order to prevent from local minimizers, we select the original estimators by meshing.Fifth, another method based on genetic algorithm is used to solve the LAD estimator. Genetic algorithm quicken the speed of searching the global superior solution.Sixth, Using all-pass model to fit and analyze Shanghai and Shenzhen stock market, we predict the future close prices respectively. It shows that the accuracy of prediction is fairly good.