Comparative Study Of Several Different Types Of Financial Time Series Models

Time series analysis is an important branch of statistics,in recent years,all kinds of time series models are widely used in various fields,especially in the financial field.AR model is one of the earliest financial time series models,and There are many methods for estimating the autoregressive parameters of AR models,and the effects of estimation are different.The GARCH models under different distributions have different effects on stock market characterization and risk value prediction.Therefore,it is of great significance to study the AR model and GARCH model In this paper,we mainly study the effect of the improved AR(1)model prediction interval in the simulation and research on parameter estimation method and prediction of AR model,At the same time,the Shenzhen composite index and Shanghai composite index are empirically studied by using different GARCH modelsFirstly,The AR(1)model of autoregressive parameters using median estimator constructs the model prediction interval,one is not with the unit root test,the other with a unit root test.Then Monte Carlo simulation experiment,by comparing sample prediction intervals of unit root tests in terms of area coverage or average length effect some are better..The AR(1)model is modeled by different parameter estimation methods,Then,different prediction methods are used to predict,and it is found that the least square estimation is good in the stationary state.Secondly,Based on the study of different GARCH rate model under Shenzhen returns can be seen there is leverage effect of the stock market,By comparison,we find that the EGARCH-t model can well describe it.At the same time using IGARCH and TGARCH model under different distribution of Shanghai stock index,TGARCH model of GED distribution can well describe the return sequence the characteristics,and then calculate the risk value of each model,then Kupiec test was carried out on the calculated VaR.Through the test,we can see that the TGARCH model under the GED distribution has the best effect on risk measurementFinally,some existing problems and future research prospects are discussed.