Data Mining Based Forecast Analysis And Research Of Financial Time Series

Following the development of computer science, artificial intelligence and statistics, data mining technology becomes possible and has gained deep development, and in the same time data mining has extended in financial time series data study, which greatly promotes the development of the new subject field—financial time series data mining. Data mining is a process of learning and gaining connotative and valuable information. As a branch of statistics, time series analysis has experienced extensive development since 60's last century. Nowadays due to the complexity of decision analysis, that traditional methods become weak in finding more complex and more covert rules and models. Based on above analysis, this paper integrates the study of data mining and financial time series.The main contents of this dissertation focus on:(1)The pretreatment of time series data , which contains original data standardization, missing data complement, noise elimination, data integration, and wavelet theory is detailed for noise elimination.(2) Financial time series econometrics models of ARMA and ARCH and related derivative models are presented, and details of determination and examination of parameters of ARMA and ARCH models are contained. Finally, based on the above analysis, Shanghai stock index is taken as an example for testing the model in forecasting analysis. Experiment showed that ARMA-ARCH model could well be used in short term forecast application.(3) Financial calendar effect is discovered, which is as the financial time series data mining application example. From the experiment, we could see after 2001, the calendar effect (month effect, week effect and holiday effect) of China financial market is distinct but different from former research result of native and abroad study.At last, the thesis is concluded, and weakness and later research directions are pointed out.