Font Size: a A A

Research On Financial Data Analytical Method Based On Fractal Technology

Posted on:2011-07-11Degree:DoctorType:Dissertation
Country:ChinaCandidate:L P NiFull Text:PDF
GTID:1119330332966798Subject:Management Science and Engineering
Abstract/Summary:PDF Full Text Request
With the rapid development of information construction of financial industry more and more data has been created. How to make effective analysis of these data becomes a key issue. In recent years according to the dynamic property,complex and nonlinearity of financial data,researchers introduce nonlinear theory to reveal the market operational rules better. Fractal technology is a branch of nonlinear theory and related researches show that fractal is a common phenomenon in financial market.The purpose of this dissertation is to do research on financial data analytical method based on fractal technology around hot and difficult questions of financial data analysis area.According to the character of financial data,fractal dimension definition,meaning and estimation methods of univariate and multivariate time series are researched;On these bases fractal dimension and data mining algorithm are combined to solve some financial data analysis problems which are similarity analysis,dimension reduction and prediction.The primary work of this dissertation includes:1. The background and significance of this dissertation are discussed;the development of the fractal theory is introduced;The principles and methods of fractal technology in analyzing financial data are summarized.2. Some commonly used fractal dimension estimation methods of financial time series are introduced;Data fitting method in late solving process of estimating fractal dimension is discussed. Least square fitting method and least square sectional fitting method are used to fitting the data respectively. The related experimental results show least square sectional fitting method can improve fitting performance and the accuracy of fractal dimension.3. In order to represent the fluctuation of financial time series better a tendency fractal dimension and its estimation method are proposed. This fractal dimension includes positive fractal dimension and negative fractal dimension. The experiments on stock data,exchange rate data and futures data demonstrate that both positive fractal dimension and negative fractal dimension indicate the uptrend or downtrend of financial market better than traditional fractal dimension indicates.4. The similarity analysis methods of financial time series are researched. A method combined with the tendency fractal dimension and K-means algorithm is proposed, and then stock index series similarity clustering is researched with this proposed method. In this method tendency fractal dimension is firstly used to represent the time series and then K-means algorithm is used to cluster the different index series. By comparison with the method that using traditional fractal dimension and K-means algorithm to cluster the stock index series, the results show tendency fractal dimension has the advantages of accurate and delicate description capability. Experimental results further demonstrate the meaning and function of tendency fractal dimension.5. Fractal dimension estimation methods of multivariate time series are analyzed and compared .An extended method is proposed to estimating the fractal dimension of multivariate time series. The fractal estimation method is simple, convenient and feasible and can get proper result.6. The feature selection method based on Ant Colony algorithm and fractal dimension is proposed for the problem of feature reduction of multivariate time series. Forecasting problem of multivariate financial time series is researched on the basis of this feature selection method. Experimental results show this improved feature selection algorithm can recognize determinant attributes and improve the accuracy of forecasting.
Keywords/Search Tags:Fractal, Tendency Fractal Dimension, Univariate Time Series, Multivariate Time Series, Similarity Analysis, Feature Selection, Time Series Prediction
PDF Full Text Request
Related items